Search

• An active deep sensing method for efficient occlusion avoidance was established. • Datasets of occlusion state to confidence and occlusion distribution were obtained. • The data correlation was learned through a multi-layer convolutional neural network. • The results proved the method's superiority in recognition efficiency and accuracy. • The active sensing method accelerates the practical deployment of harvesting robots. With the increasing shortage of agricultural labor, the development of harvesting robots is becoming more and more urgent. Most of them require vision to locate the target, however, occlusion is common in agricultural environment, which restricts the accuracy of visual target recognition, and even leads to failure in serious cases. The active perception method is an effective means, but how to efficiently find the best observation position remains difficult to avoid the waste of time caused by repeated invalid motion. Targeting these problems, an active deep sensing method is proposed for harvesting clustered and single fruits. First, the region of interest of the target is extracted by a segmentation network, and then the occlusion status of it is obtained by image processing methods. Taking the current observation position as the starting point, the camera is moved within a matrix to form confidence and occlusion rate distribution maps. After establishing a series of occlusion rate and confidence matrix datasets, a designed deep network has been trained, which is used to predict the maximum confidence/minimum occlusion rate position after the current occlusion status is estimated. To verify the reliability of the method, laboratory and field experiments were carried out for apples and clustered tomatoes. After 1000 times of verification, results show that the successful pick/recognition rate is increased by 38.7 %, and the average successful recognition time is 5.2 s, which is 63.1 % and 46.4 % faster than that of a fixed movement method and a simple heuristic method. [ABSTRACT FROM AUTHOR]

• Developed automatic maize seeding machine was tested for watering system. • Comprehensive watering spray angle performance index was proposed. • Depth coupled with color was used for accurate image segmentation of sand. • A medium 55° spray angle is recommended for practical applications. Seeding is a significant operation in maize cultivation, which is mostly manual, so the germination rate is key information. We developed an automatic seeding machine that performs sand filling, seed placement, watering, and seed covering, where watering is a critical operation. Spray angle, a key factor influencing the watering process and seed germination characteristics, was determined using image analysis. The current system's spray angle includes three settings (35°, 55°, 75°), and its performance was evaluated in terms of watering area ratio (WAR, small watering area can result in insufficient moisture and wetting area for the seeds), seed pattern distortion (SPD, irregular seed pattern would interfere with the detection of germinated seedlings), sand surface deformation (SSD, a measure of moisture content in sand due to its surface undulations), and water spraying efficiency (WSE, a small value indicates poor system performance). For WAR, SPD, and SSD evaluation, image data of a tray loaded with sand and planted with seeds using RGB-D (color [RGB] coupled with depth [D]) were collected before and after watering, followed by data pre-processing (general image segmentation). An empty tray was used to evaluate the WSE by measuring its mass before and after watering. For small (35°), medium (55°), and large (75°) spray angles, the WAR was 87.2 %, 96.4 %, and 99.3 %; the average seed displacement was 20.0, 14.7, and 3.1 mm/tray; the average volume of the puddle was 11.3, 0, and 0 cm3; and the WSE was 9.49, 8.89, and 5.13 g/s, respectively. For a comprehensive and objective comparison of the three spray angles, the spray angle performance index (SAPI) was proposed, and the corresponding values of SAPI for spray angles were 0.11, 0.47, and 0.39, respectively. It is concluded that the medium spray angle (55°) generates the most satisfactory performance and should be applied in practical applications. Furthermore, the developed workflow for systematically evaluating the sprayer's performance has the potential to be used as an automatic and objective evaluation method for various spray nozzles and angle settings from the automatic seeding and germination standpoint. [ABSTRACT FROM AUTHOR]

• Propose a grass vegetation height detection framework model called ASHS. • Propose a novel method for estimating the height of Achnatherum splendens based on image processing. • Confirm that small quadrat grassland (<1m2) is suitable for single batch testing. • Build a complete set of Achnatherum splendens object detection and image segmentation datasets. The growth of Achnatherum splendens affects the growth of dominant forage grasses in grasslands, which leads to a decrease in grassland biomass and an increase in grassland ecological environment degradation. It is a time and labor-consuming work to measure the height of Achnatherum splendens in wild grassland, so developing an intelligent grass height estimation method is indispensable. This study aims to propose a novel method for estimating the height of Achnatherum splendens based on image processing. The mobile ground robot with the binocular depth camera captures image information from different angles. This method acquires the edge features of the image from different angles and obtains the height pixel points by feature matching and clustering. The pixel points filtered by deep learning models' Bounding Box are converted to the world coordinate system using spatial geometry. This study took YOLOv7-X as an example, as its average precision (AP) detection rate was 96.6 %, which was superior to other deep learning algorithms in terms of overall performance. In the mean height estimation experiment of 10 test sites, the mean relative error (MRE) was 6.4 % and coefficient of determination (R2) was 0.74. In the mean height estimation experiment of 10 test sites, the MRE was 3.1 % and the R2 was 0.85. It is proven that this proposed method is effective and strongly agrees with actual Achnatherum splendens heights measured manually. Experiments also showed that this method performed better in the small sample test sites (The basal coverage was no more than 1 m2 in this study). This study encapsulated the above methods and proposed a model called ASHS, which could estimate the height of grassland vegetation represented by Achnatherum splendens. Different algorithm modules in the model can be replaced according to the vegetation situation. This study may contribute to estimating the quadrat height of grasses and developing intelligent robots in grassland resource surveys. [ABSTRACT FROM AUTHOR]

• MSBANet achieves higher discrimination accuracy of tobacco leaf maturity than GhostNet. • MSBANet is more practical to identify the maturity of tobacco leaves in natural environments. • MSBANet based on "plant phenotype + computer" shows a potential for development of tobacco harvesting robots. Traditional machine vision methods are susceptible to environmental influences, low recognition accuracy, and poor real-time performance in identifying tobacco leaf maturity. Currently, the maturity of tobacco leaves heavily relies on manual observation, and existing research on tobacco maturity discrimination focuses on standard light sources and standard background, which difficult to be applied to tobacco harvesting robots in natural environment. To tackle this problem, a maturity discrimination method based on the branch attention mechanism is proposed. Firstly, the region of interest in the extracted tobacco leaf images is obtained as a foundation for subsequent image processing. Then, a deep learning network model called the Multi-Scale Branch Attention Network (MSBANet) is designed for identification of tobacco leaf maturity. The effectiveness of MSBANet stems from its newly designed modules for multi-scale information extraction and branch attention. These modules help overcome the impact of lighting conditions and enable the extraction of more comprehensive and informative feature representations, capturing the relationship between phenotype and maturity. On the self-constructed dataset of tobacco leaf maturity (color images are 576 × 768 pixels), MSBANet-L achieves 89.6 % accuracy with 7.1GFLOPs and MSBANet-S achieves 88.2 % accuracy with 1.7 GFLOPs. When compared with widely used BP and MobileNet, as well as state-of-the-art structures such as YOLOv7-tiny-SiLU and FasterNet-T0, MSBANet demonstrates excellent performance. This method preliminarily explores the maturity discrimination of tobacco leaves in actual environments and shows a potential for development of tobacco harvesting robots. [ABSTRACT FROM AUTHOR]

• UAVs are widely used in plant disease diagnosis and monitoring. • Wheat diseases such as rusts and blights are accurately identified using UAVs. • An optimal trained model is necessary to increase the accuracy of disease identification in wheat. • UAVs can be a valuable instruments for acquiring detailed information on plant diseases with exceptional spatial, temporal, and spectral precision. Wheat is a major grain crop cultivated worldwide, and its production is influenced by various diseases and pests, leading to significant production losses. Timely disease detection in wheat is crucial for farmers to apply effective control measures and prevent disease spread and potential reduction of yield and quality. However, conventional disease detection typically demands trained professionals and sophisticated laboratory equipment for accurate identification, making it economically unfeasible for large-scale wheat cultivation. In recent years, various research efforts have focused on finding alternative methods for disease detection in wheat using Unmanned Aerial Vehicles (UAVs) in conjunction with deep learning and image processing techniques. Nevertheless, there is a notable absence of comprehensive studies that review and compare these various approaches. Our article seeks to bridge this gap by providing a comparative analysis of different types of UAVs, sensors, image processing methods, and classification techniques employed in wheat disease detection. Additionally, we delve into the opportunities and challenges associated with the use of UAVs as tools for disease monitoring in wheat cultivation. Our review shows the potential of UAVs for automated disease detection and monitoring in wheat. [ABSTRACT FROM AUTHOR]

• The effect of broken kernel content on pore structure parameters was studied. • The porosity and permeability in the grain pile decrease with increase in the broken kernel content. • The small pores (diameter < 1 mm) and isolated pores increase with increase in the broken kernel content. • The pore throat size of grain piles decreases and the tortuosity increases with increase in the broken kernel content. • The average coordination number of pores increases with increase in the broken kernel content. The airflow in grain piles is affected by the content of broken grains, which causes changes in the pore structure within the grain pile. To investigate the changes in structural parameters at different broken grain contents, CT scanning and image processing were used to study the pore structure characteristics of broken maize grain contents of 0 %, 4 %, 8 % and 15 % respectively. The airflow distribution in the grain pile under different broken grain contents was studied by numerical simulation. The results showed that as the broken kernel content increased from 0 to 15 %, grain pile porosity gradually decreased, with a change rate of 6.97 %. An increase in the content of broken grains causes a decrease in the diameter and volume of the pores in the grain piles. The average value of the pore coordination number in the grain pile decreased as the broken grain content increased. When the content of broken grains is 0–15 %, the rates of change of pore diameter and tortuosity in the grain piles are 14.94 % and 24.94 %, respectively. As the broken grain content increased, the length and diameter of the pore throat in the grain pile decreased, while the fractal dimension and tortuosity increased. The increased broken grain content in the grain piles impedes the flow of air through the pores, thus affecting ventilation efficiency. [ABSTRACT FROM AUTHOR]

Plant diseases have a significant impact on both the quantity and quality of crop yields. Therefore, the timely detection of these diseases is crucial for facilitating early treatment. Given that most plant diseases affect leaves and fruits, visually inspecting plants for disease symptoms has become increasingly vital to minimize potential damage. In the field of precision agriculture, achieving accurate and rapid automated identification of plant diseases has become essential. To address this challenge, various computer vision and deep learning-based models have been employed. Thanks to their impressive performance, deep learning has emerged as the preferred method for plant disease detection. In this context, we present an effective plant disease identification framework based on a transformer structure, designed to capture long-range features. Additionally, we suggest enhancements such as BatchFormerV2, the layer-wise adaptive moments optimizer for batch training (LAMB), and complete intersection over union (CIoU) loss to the original model. These additions enrich the model's ability to learn fine-grained features and stabilize the training process. Importantly, the proposed model is interpretable, as demonstrated by the analysis of attention weights to make the prediction process transparent and comprehensible. The proposed model outperforms previous convolutional and vision transformer-based models. The model achieves the highest mAP value of 56.3 on the large-scale plant disease dataset used in this study. • A huge plant disease dataset for 6 species that contains over 123,000 images. • An efficient transformer-based plant disease detection framework. • Analysis of the localized disease region using the transformer's self-attention weights. • The proposed model outperformed previous state-of-the-art object detection models. [ABSTRACT FROM AUTHOR]

This study presents an automated lameness detection system that uses deep-learning image processing techniques to extract multiple locomotion traits associated with lameness. Using the T-LEAP pose estimation model, the motion of nine keypoints was extracted from videos of walking cows. The videos were recorded outdoors, with varying illumination conditions, and T-LEAP extracted 99.6% of correct keypoints. The trajectories of the keypoints were then used to compute six locomotion traits: back posture measurement, head bobbing, tracking distance, stride length, stance duration, and swing duration. The three most important traits were back posture measurement, head bobbing, and tracking distance. For the ground truth, we showed that a thoughtful merging of the scores of the observers could improve intra-observer reliability and agreement. We showed that including multiple locomotion traits improves the classification accuracy from 76.6% with only one trait to 79.9% with the three most important traits and to 80.1% with all six locomotion traits. • T-LEAP was used to track keypoints on the hooves, head and back of the cows. • Six locomotion traits associated with lameness were computed from the keypoints. • Lameness classification improves when including multiple locomotion traits. [ABSTRACT FROM AUTHOR]

• This study presents an extensive systematic literature review on WD recognition. • A set of 94 research articles published between 2005–2022 has been selected for analysis. • A framework has been proposed to classify the research articles into four categories. Due to wheat diseases (WD), the global rate of wheat production is decreasing by 3.6% annually. With the help of computer vision technology, WD recognition is not a challenging task but motivates the concepts of image processing, image segmentation, feature extraction, and AI-based recognition models. The objective of this study is to review and systematically analyze studies that have been published between 2005 and 2022. The authors make an effort to determine the important developments in image segmentation models, tools, datasets, and comparative analysis for the accuracy of the recognition model. The current study follows the standard systematic literature review (SLR) approach and selects 638 studies from five different web source databases. Among 638 studies, 544 studies were discarded in the study extraction process. A total number of 94 studies have been published in 45 reputed journals, and 49 conferences that have been identified with evaluation, validation, proposed, and philosophical criteria. After analysis, ten types of image segmentation models were identified. The most prominent clustering-based image segmentation technique (34.78%) is used for powdery mildew and stripe rust WD recognition. During WD recognition, the accuracy performance parameter is found to be most prominent. China and India are the two countries on the Asian continent that contribute to WD recognition. The current study summarizes the findings of WD research and highlights the need for standard datasets and accuracy. It highlights the importance of exploring and developing more precise and hybrid segmentation classification models for WD recognition. [ABSTRACT FROM AUTHOR]

• CattNISDB, a new public dataset for cattle, facilitates research. • CattNIS proposes an image-based approach for cattle recognition using retinal vascular patterns. • An automated GUI identifies animals through retinal images. • The CattNIS system, with 300 animals and 2430 images, facilitates accurate and realistic research findings. Small and large farm animals are frequently identified using techniques like microchipping or tagging. However, these techniques may have a detrimental impact on animal welfare and may result in theft and fraud. Modern and sophisticated identification procedures using biometrics, including retina, face, and iris recognition, are available and have less of a detrimental effect on animal welfare. Because biometric markers are unchangeable and unaffected, it is feasible to stop situations of theft and fraud. This study uses retina images to identify animals based on each animal's different and specific retinal patterns. To the best of our knowledge, there isn't a publicly accessible dataset for images of animal retinas. The three main goals of this study are to: 1) produce a novel dataset of Cattle retinal images and make it publicly available; 2) create a successful image processing-based system for retinal identification and recognition; and 3) create a graphical user interface (GUI) that enables querying the database to ascertain whether a specific Cattle retina is present. To accomplish these goals, a dataset comprising 2430 images from the left and right eyes of 300 cattle was produced. After the dataset was created, the animal retinal images were segmented using image processing methods such as scaling, color transformations, image sharpening, contrast enhancements, noise filtering, and histogram equalization. By using the segmented images with the SURF, FAST, BRISK, and HARRIS techniques, distinctive retinal features were found. During the identification phase, the animal whose retinal characteristic matched the sought animal's retina the closest was recognized. Within the developed CattNIS system, the SURF approach has the highest accuracy rate of 92.25 percent. The results of this study show that the identification of cattle using retinal vascular patterns is highly successful. The availability of the collected cattle retinal images from this study is expected to significantly contribute to the advancement of future research in this field, both in terms of quantity and quality. [ABSTRACT FROM AUTHOR]

• Open-source dataset: Five categories of dataset exported using multiple augmentation techniques. • Single-stage model that achieves weed detection and segmentation. • Integrating C3x module within the backbone for detailed feature extraction. • Model trained on C4 category achieved the best detection and segmentation scores of 85.4 % and 82.1 %, respectively. • Model with ONNX format gained 1.25x inference speed on an edge device. Deep learning (DL) inspired models have achieved tremendous success in locating target weed species through bounding-box approach (single-stage models) or pixel-wise semantic segmentation (two-stage models), but not both. Therefore, the goal of this research study was to develop a single-stage DL architecture that not only locate weed presence through bounding-boxes but also achieves pixel-wise instance segmentation on unmanned aerial system (UAS) acquired remote sensing images. Moreover, the developed architecture experiments on integrating a novel C3 and C3x module within its backbone for dense feature extraction, as well as ProtoNet (Prototypical network) in its head component for weed masking. Furthermore, the proposed architecture has been trained on five categories of dataset exported using multiple combinations of various dataset augmentation techniques, namely, C1, C2, C3, C4, and C5, for which multiple metrics were assessed on desktop graphical processing unit (GPU) and a palm-sized edge device (AGX Xavier). Results suggest that category C4, a combination of six data augmentation techniques, outperformed the remaining categories by achieving precision scores of 85.4 % (bounding-boxes) and 82.8 % (masking) on a GPU. Whereas, the same model converted to TorchScript was able to achieve 79.1 % and 77 % bounding-box and masking accuracy on an edge device, respectively. The model developed in this research has two potential applications when integrated with site-specific weed management technologies. First, it enables real-time weed detection, allowing for the immediate identification of weeds for spot-spraying applications. Second, it facilitates instance weed masking, aiding in the estimation of weed growth extent in actual field conditions. Moreover, the developed architecture combines both computer vision applications - detection and instance segmentation – to provide comprehensive information about weed growth, eliminating the need for multiple algorithm. [ABSTRACT FROM AUTHOR]

This study introduces a lightweight visual segmentation model named TinySegformer, specifically designed for agricultural pest detection, aiming to address edge computing issues in real-world scenarios. The innovation of TinySegformer lies in its effective combination of Transformers and neural networks, enhancing its ability to handle semantic segmentation tasks and significantly improving image segmentation performance. Additionally, through the use of sparse attention mechanisms and quantization techniques, it adopts a lightweight architecture, enabling the model to adapt to the computational and storage limitations of edge devices. This study evaluates TinySegformer on multiple datasets, revealing that whether on public datasets or our self-collected datasets, TinySegformer outperforms current mainstream visual segmentation models such as DeepLab, SegNet, UNet, PSPNet (Pyramid Scene Parsing Network), FCN (Fully Convolutional Networks), etc. In terms of key performance indicators like precision, recall, accuracy, mIoU, Dice coefficient, and FPS, TinySegformer shows outstanding results, reaching 0.92, 0.90, 0.93, 0.85, 0.91, and 65 respectively, and achieves real-time image processing at 32.7 frames per second on edge devices. Furthermore, the study elaborately discusses the process of deploying TinySegformer onto NVIDIA Jetson devices, and successfully adapts the model to resource-constrained devices through network pruning and quantization techniques. In conclusion, with its efficiency, accuracy, and lightweight characteristics, the TinySegformer model provides a robust and practical solution for agricultural pest detection, offering new insights and directions for future research in the field of agricultural pest monitoring. • A dataset with 102 pest types enables precise pest detection and actionable insights in real-world applications. • TinySegformer's combination of Transformer and neural network technologies enhances image segmentation accuracy. • Its lightweight architecture is tailored for edge computing, ensuring precise pest detection in complex environments. • The model's performance is thoroughly assessed using diverse metrics like accuracy, recall, precision, mIoU, and FPS. • Adapted for mobile devices through network pruning and quantization, it achieves real-time image processing at 32.7 FPS. [ABSTRACT FROM AUTHOR]

• A novel Optical Droplet Edge Imaging method for droplet measurement is proposed. • A fast in-situ measuring portable device for droplet deposition is designed. • An image processing algorithm is proposed to obtain deposition performance indicator. • The droplet deposited area to volume is calibrated with average deviation 3.22%. • This device outperforms the water sensitive paper in terms of ease of use and digitization. Accurate and rapid measurement of droplet deposition on the spray target provides feedback data on spray quality to optimise application parameters and confirm crop protection efficacy. However, simultaneously meeting the requirements for field measurement remains a challenge. In this paper, a novel Optical Droplet Edge Imaging (ODEI) method based on a portable droplet deposition and distribution measurement device is proposed to enable in-situ reusable sampling and on-line automatic digitisation for field spraying. First, the principle of the ODEI method is presented to realise the sampling state closer to droplet deposited on the leaves. Then, a low-cost portable in-situ droplet deposition measurement device is designed and the optimal light field parameters of the device are determined. Finally, the simultaneous acquisition of the top and bottom side of the droplet deposition image of the collector is obtained and its image processing algorithm is presented to obtain the performance indexes of droplet deposition, such as individual droplet size and distribution, total droplet number, volume median diameter, droplet density and the ratio of droplet amount between the top and bottom side of the collector. The relationship between droplet volume and image area is calibrated with an average deviation of 3.22 %. Experimental results show that the VMD deviation between the device and the WSP image measured by same image processing method is 15.34 μm. Due to the difference in deposition state between the device and the water sensitive paper (WSP), the droplet deposition area conversion from this device to the WSP was also calibrated. The absolute deviation between the WSP moving average coverage rate and the converted collector coverage rate of this device is 1.21 %. The measurement result of the spray experiment shows that this device has the same measurement capability as the WSP, as the droplet diameter distribution is very consistent with the WSP distribution. This in-situ and automatic measurement method is a promising and useful addition to the spray performance monitoring tool for field droplet deposition. [ABSTRACT FROM AUTHOR]

Consistency of crop establishment is a measure of uniformity of crop attributes, such as plant stand count, crop emergence rate, and plant spacing across the field. Quantifying consistency during the early crop growth stage is important for establishment decisions to use targeted nutrients and to facilitate timely replanting in inconsistent crop regions. Crop consistency can be analysed using two key parameters: plant stand count and spacing statistics since they provide insight into plant density and its emergence percentage. However, manual assessment of them is time-consuming, prone to errors, and labour-intensive in large fields. An alternative method is proposed to automate estimating these parameters using field imagery under uncontrolled settings. We use the YOLOv5-based object detection model for plant counting, which attains a mean average precision of 0.956 to detect Canola plants. A Lightweight U-Net model is proposed to segment rows, followed by Guo–Hall thinning and Probabilistic Hough Transform to determine inter-row and inter-plant spacing. Our proposed row segmentation model achieves a mean Intersection over Union (mIoU) of 0.8444 with class-wise IoU of 0.9925 and 0.6963 for background and crop using fewer parameters. The new architecture uses only 14M parameters and achieves performance comparable to the state-of-the-art U-Net (32.5M) and SegNet (29M). • Develops a novel method for quantification of Canola crop consistency. • Proposes a lightweight U-Net based network for segmenting crop rows. • Evaluation of one-stage object detectors for plant stands counting in RGB images. • Evaluation of the method on multiple crop stages under variable weed pressure. [ABSTRACT FROM AUTHOR]

• We propose Fast-Unet based on Unet pruning and optimization to recognize navigation paths between rows of peach, orange and kiwifruit. • We propose to use ASPP for feature extraction in model down-sampling and Bilinear interpolation for up-sampling to improve model inference speed and segmentation accuracy. • We adopt the loss function Dice Loss that can optimize the intersection of union and the BCE Loss that can improve the accuracy of difficult classification categories as the loss function of the model. • We use transfer learning to transfer models trained on a large sample peach dataset to small sample datasets of orange and kiwifruit, verify the feasibility of expanding the use of transfer learning through small samples, and propose a generalized method for navigation path recognition between rows of fruit trees. • We extract the path boundaries using the Canny operator and fit the navigation lines using the least square method. The navigation path recognition has been recognized as one of the most important subtasks of intelligent agricultural equipment in orchard operations. However, there are still some challenges in recognizing navigation paths between rows of fruit trees, including the accuracy, real-time performance, generalization of deep learning models. The Fast-Unet model was proposed by pruning and optimization based on Unet for recognizing navigation paths between rows of fruit trees, which inherited encoding–decoding structure and multi-layer feature sensing capability. The number of convolutional kernels used to extract features in the Fast-Unet was reduced to one-fourth of that in Unet to improve inference speed. To address the blurring of the boundary of the recognized object due to the reduction in the number of convolutional kernels, the atrous spatial pyramid pooling (ASPP) module was used in the encoding part to extract the multiscale information to improve the recognition accuracy. The navigation path edges determined by Fast-Unet and Canny operators, navigation lines and yaw angles were generated by the least square method.. In this study, the Fast-Unet model was first trained on the peach dataset, and then the trained model was transferred to the small dataset of oranges and kiwifruits for navigation path recognition to verify the generalization. The Mean Intersection over Union (MIOU) of the Fast-Unet for peaches, oranges and kiwifruits navigation path extraction accuracy were 0.977, 0.987 and 0.956, respectively. The mean difference between the predicted yaw angle of peaches, oranges and kiwifruits navigation paths and the labelled were 0.397°, 0.102° and 0.239°, respectively. In terms of real-time performance, the inference speed was 48.8 frames per second (FPS) to process the RGB image data on a single-core CPU. The inference speed of the Fast-Unet model was 1.59 times higher than that of Unet. Through transfer learning, the Fast-Unet model can realize real-time recognition of navigation paths for peaches, oranges and kiwifruits. These results can provide technical and theoretical support to the development of navigation equipment and visual slip prediction for intelligent agricultural machinery in the orchard. [ABSTRACT FROM AUTHOR]

• A novel DFN-PSAN model is proposed to identify crop diseases in natural farmland environments accurately. • The multi-level Deep Information Feature Fusion Network (DFN), which can effectively extract and fuse relevant features from different network layers, improves the localization of infected plant disease areas. • Pyramid Squeeze Attention (PSA) fuses contextual information at different scales and produces better pixel-level attention. • The construction of the PSA attention classification network (PSAN) can effectively utilize the important feature information of DFN to achieve competitive performance on complex disease symptom datasets in the field. • The t-SNE and SHAP methods enhance the transparency of the model in terms of feature clustering and discrimination of multi-category disease attention, respectively. Accurate identification of crop diseases is an effective way to promote the development of intelligent and modernized agricultural production, as well as to reduce the use of pesticides and improve crop yield and quality. Deep learning methods have achieved better performance in classifying input plant disease images. However, many plant disease datasets are often constructed from controlled scenarios, and these deep learning models may not perform well when tested in real-world agricultural environments, highlighting the challenges of transitioning to natural farm environments under the new demand paradigm of Agri 4.0. Based on the above reasons, this work proposes using a multi-level deep information feature fusion extraction network (DFN-PSAN) to achieve plant disease classification in natural field environments. DFN-PSAN adopts the YOLOv5 Backbone and Neck network as the base structure DFN and uses pyramidal squeezed attention (PSA) combined with multiple convolutional layers to design a novel classification network PSAN, which fuses and processes the multi-level depth information features output from DFN and highlights the critical regions of plant disease images with the help of pixel-level attention provided by PSA, thus realizing effective classification of multiple fine-grained plant diseases. The proposed DFN-PSAN was trained and tested on three plant disease datasets. The average accuracy and F1-score exceeded 95.27%. The PSA attention mechanism saved 26% of model parameters, achieving a competitive performance among existing related methods. In addition, this work effectively enhances the transparency of the features of the model attention to plant diseases through t-SNE with SHAP interpretable methods. [ABSTRACT FROM AUTHOR]

Large-scale animal husbandry currently serves as the primary source of protein and essential nutrients for human consumption. Ensuring adequate access to animal weight has emerged as a pivotal approach for monitoring animal health and production performance. However, with a plethora of weighing methods available, the need to compare these diverse technologies according to the specific requirements of various animal species has become a pressing concern. Addressing this issue is imperative for the advancement of weighing technology. Consequently, we have undertaken a comprehensive literature review that delves into the developmental trajectory of weighing technology and provides in-depth descriptions of its corresponding characteristics. We aim to discern the technical challenges encountered at different stages of development and offer a thorough insight into the strategies employed to overcome these challenges. In our investigation, we have amalgamated evaluation criteria from various weighing methods into this developmental lineage. Our primary works are as follows: First, we compiled a comprehensive collection of animal weight application scenarios, encompassing species such as pigs, cattle, ducks, broilers, and fish. We systematically organized weight measurement methods proposed by researchers tailored to the unique requirements of each scenario. Secondly, we delved into the evolution of animal weighing techniques, considering the inherent limitations of direct weighing methods. We meticulously cataloged feature parameter extraction methods from various studies within the traditional image-based weighing domain and conducted an in-depth comparative analysis of not only the weighing methodologies themselves but also their associated evaluation metrics. Third, our focus shifted to addressing the deficiencies of traditional image-based weighing techniques. We honed in on deep learning-based approaches, systematically dissecting the critical components of image acquisition, pre-processing, model training, and evaluation. In this examination, we scrutinized the vulnerabilities and potential remedies associated with image quality's impact on the precision of animal weight estimation. Lastly, in light of the current state of research, we explored the challenges that persist in the realm of deep learning-based animal weighing. Within this context, we put forth prospective solutions to overcome these obstacles. We hope this review will provide a clearer understanding of the evolution of weighing methods and explore potential solutions to the challenges faced in deep learning weighing methods currently. This will, in turn, foster the advancement of weighing systems for various applications. • Compiled animal weighing methods for various animal types and different body sizes. • The first description of animal weighing based on deep learning. • Overview of the content of datasets for image-based animal weighing. • Summarizes the methods of weighing different weight types of animals. • Explore challenges in deep learning for animal weighing and propose solutions. [ABSTRACT FROM AUTHOR]

[Display omitted] • We developed AICropCAM, an edge-computing enabled camera system for crop monitoring. • It integrated Raspberry Pi and Arduino for image processing and LoRa communication. • It ran a stack of four deep neural networks to characterize multiple plant/canopy parameters. • We quantified the power consumption and speed of the system for edge image-processing. • AICropCAM is a next-generation enabling technology for real-time, low-latency Ag applications. Precision Agriculture (PA) promises to meet the future demands for food, feed, fiber, and fuel while keeping their production sustainable and environmentally friendly. PA relies heavily on sensing technologies to inform site-specific decision supports for planting, irrigation, fertilization, spraying, and harvesting. Traditional point-based sensors enjoy small data sizes but are limited in their capacity to measure plant and canopy parameters. On the other hand, imaging sensors can be powerful in measuring a wide range of these parameters, especially when coupled with Artificial Intelligence. The challenge, however, is the lack of computing, electric power, and connectivity infrastructure in agricultural fields, preventing the full utilization of imaging sensors. This paper reported AICropCAM, a field-deployable imaging framework that integrated edge image processing, Internet of Things (IoT), and LoRaWAN for low-power, long-range communication. The core component of AICropCAM is a stack of four Deep Convolutional Neural Networks (DCNN) models running sequentially: CropClassiNet for crop type classification, CanopySegNet for canopy cover quantification, PlantCountNet for plant and weed counting, and InsectNet for insect identification. These DCNN models were trained and tested with >43,000 field crop images collected offline. AICropCAM was embodied on a distributed wireless sensor network with its sensor node consisting of an RGB camera for image acquisition, a Raspberry Pi 4B single-board computer for edge image processing, and an Arduino MKR1310 for LoRa communication and power management. Our testing showed that the time to run the DCNN models ranged from 0.20 s for InsectNet to 20.20 s for CanopySegNet, and power consumption ranged from 3.68 W for InsectNet to 5.83 W for CanopySegNet. The classification model CropClassiNet reported 94.5 % accuracy, and the segmentation model CanopySegNet reported 92.83 % accuracy. The two object detection models PlantCountNet and InsectNet reported mean average precision of 0.69 and 0.02 for the test images. Predictions from the DCNN models were transmitted to the ThingSpeak IoT platform for visualization and analytics. We concluded that AICropCAM successfully implemented image processing on the edge, drastically reduced the amount of data being transmitted, and could satisfy the real-time need for decision-making in PA. AICropCAM can be deployed on moving platforms such as center pivots or drones to increase its spatial coverage and resolution to support crop monitoring and field operations. [ABSTRACT FROM AUTHOR]

• Improved UNet network enhances maize crop row segmentation. • Adaptive perspective transformation corrects image distortions. • Detects maize at various growth stages and under complex field conditions. • Provides technical support for precise variable-rate spraying for maize seedlings. Crop-row line detection is key to realizing the accurate spraying of crops; however, it is challenging because of uneven lighting, weeds, and missing plants within rows. In this study, we propose an automatic detection method for maize width and row centerlines based on image processing that can detect maize at various growth stages and under complex field conditions. Our method consists of three stages: (i) Maize crop-row segmentation based on ResC-UNet, an improved version of U-Net that integrates ResNet-50 and attention mechanism modules. (ii) Correction of distorted images by adaptive perspective transformation to prevent parallel crop rows from creating "vanishing points" at the top of the image. (iii) Detection of the width and missing plants in the row through width fitting, and extraction of the actual row line according to the positional relationship between the width and row line. We established a dataset of maize crop rows by capturing images of maize at different growth stages in the field. Offline experiments on this dataset showed that our method achieved 94.9% accuracy for maize row segmentation, 1.71° average error angle for row-line fitting, and 91.51% average pixel accuracy for width detection. It thus enables technical support for precise variable-rate spraying along rows during the maize seedling stage. [ABSTRACT FROM AUTHOR]

• A set of processing procedure for images captured by UAV is designed. • The improved U-net is used for semantic segmentation to screen the effective region. • The improved VGG-19 is used to locate and identify the disease on the target leaf. • The testing phase is performed under a testing set of different origin from the training set. In order to realize the acquisition of plant leaf images by agricultural drones and perform regional segmentation and disease recognition on leaf cluster images over a wide area, this paper takes grape leaves as an example and designs a set of processing procedures combining the improved U-net and VGG-19 networks. This paper aims to address the difficulty of obtaining target leaves for recognition in complex environments due to the presence of multiple invalid leaves. First, the feature extraction process is performed on the training and validation datasets to reduce the input parameters of the network and increase the training speed. Second, the test data is divided into three datasets: low-light, jitter interference, and two-factor combination. After recovery processing, the Multi-fusion U-net network is fed to locate diseased leaves. Finally, the improved VGG-19 network was used again to locate and identify the disease. Experimental results show that the proposed procedure achieves satisfactory performance in UAV image processing. The average accuracy of segmentation reaches 71.91%, and the identification rate of disease location is increased by 12.33% after segmentation, which provides a strong practical basis for the implementation of unmanned smart ecological farms. [ABSTRACT FROM AUTHOR]

• A new initialization scheme was proposed to optimize the performance of unsupervised learning algorithms. • Image segmentation was improved by incorporating supreme color features to the classification algorithm. • All clustering algorithms were proved to be robust under a complex field environment. The variability of illumination and weather conditions lead to a big challenge for infield image segmentation. Therefore, robust, fast, and automated algorithms are highly required to obtain reliable image segmentation results. This research was aimed to develop efficient unsupervised clustering algorithms for oilseed rape image segmentation in the field. The Naïve Bayes rule was first employed to select a supreme color feature from ten color models. An initialization approach based on the genetic algorithm (GA) was then used to define the initial cluster centroids for subsequent Gaussian mixture model (GMM), self-organizing map (SOM), fuzzy c-mean (FCM), and k-means algorithms. The length of the chromosome was determined using cluster validity indices. Finally, the performances of these algorithms were evaluated based on the image segmentation quality and computation time. After testing the proposed method on the image datasets from two fields, the results revealed that the highest segmentation accuracy of 96% was obtained using the optimized SOM and the lowest computation time was obtained using the k-means. The GA-based initialization speeded up the convergence process and ensured consistent labeling between runs. All clustering algorithms were proved to be robust to varying illumination conditions and can process images with a very complex background in an automated fashion. [ABSTRACT FROM AUTHOR]

• The image processing method is effective for obtaining the area of rice grain. • The rice yield can be reliably predicted from panicle grain area. • "Five-point calibration model" could achieve rapid yield predictions in the field. Rice yield measurement in paddy fields is of great importance for regulating the balance of national grain supply and demand, but it is laborious and time consuming. In this paper, a fast yield measurement method based on rice panicle attributes and modelling is presented. To develop the method, first, the panicle attributes that are the main determinants of grain weight were identified through regression analysis. Second, the correlations between grain area and weight parameters were established and verified for different cultivars. Finally, a "five-point calibration model" was developed to rapidly estimate the weight parameters, and the estimated values and actual yields were compared. The results showed that the image processing method was capable of extracting the area of grain per panicle more accurately than other methods and that it was suitable for different cultivars: the lowest accurate extraction rate was 90.50%. All the coefficient of determination (R2) values of the weight prediction models of rice panicle grain area and the weight parameters were greater than 0.80, and the highest R2 value was 0.96. The R2 values of the "five-point calibration model" of grain area and grain weight were all greater than 0.99. When the predicted yields calculated by the "five-point calibration model" were compared with the actual yields, the relative errors ranged from 1.36% to 8.64%; i.e., they were all less than 10%. These results indicated that rice panicle 2D image modelling could enable rapid yield predictions in the field. [ABSTRACT FROM AUTHOR]

• A robust image processing algorithm is applied in trout fish processing machine in which trout fin position is used as system functioning data. • Testifying fin segmentation process ensures the selected color channel performance. • A safety algorithm was defined based on the processed images before any functioning of the processing subsystems. • Machine vision facilitates processing the variable trout fish. • Using this algorithm results in an automatic fin detection by machine vision which no longer requires any classification with artificial intelligent like neural networks. Fish processing in small and medium fish supplying centers requires an intelligent system to operate on different sizes. Therefore, an image processing algorithm was developed to extract the proper head and belly cutting points according to the trout dimensions. The algorithm detects the fish orientation and location of pectoral, anal, pelvic, and caudal fins. In this study, each of the trout images was divided into slices along its length in order to segment the fins and extract cutting points. The channel 'B' of RGB color space was considered in both initial segmentation and fin detection stages among the examined channels of RGB, HSV, and L*a*b* color spaces. The back-belly and head-tail sides were detected with an accuracy of 100% based on gray intensity values and head to tail ratio, respectively. Furthermore, performing an analysis of variance (ANOVA) resulted in an F-value of 64.82 among the fins. Conducting a t -test among the mean intensity values of the fins and non-fin regions of channel 'B' resulted in the highest distinction with t-values of 90.30, 78.07, 74.28, and 86.01 with p < 0.01 for the pectoral, pelvic, anal, and caudal fins paired with the corresponding non-fin region, respectively. The results showed that the selected 'B' channel is the adequate one for fin segmentation. The fin detection process showed an overall sensitivity, specificity, and accuracy of 86.05%, 99.97%, and 99.87%, respectively. By solving the line determination error in 8.24% and the extra object error in 4.12% of the samples, the overall fin identification accuracy was 100%. Finally, after extracting the fin regions, the start point of the pectoral fin and the end point of the anal fin will be applied in the trout processing system as the head and belly cutting points, respectively. [ABSTRACT FROM AUTHOR]

• Literature on detection/classification of plant nutrient deficiencies was surveyed. • Focus was on machine learning techniques using proximal images. • RGB, multispectral, hyperspectral and chlorophyll fluorescence sensors were considered. • The use of regression and classification models was detailed and discussed. • Main challenges and possible solutions were addressed in detail. During the last decade, the combination of digital images and machine learning techniques for tackling agricultural problems has been one of the most explored elements of digital farming. In the specific case of proximal images, most efforts have been directed to the detection and classification of plant diseases and crop-damaging pests. Important progress has also been made on the use of close-range images to determine vegetal nutrient status, but because such studies are fewer and more scattered, it is difficult to draw a complete picture on the state of art of this type of research. In this context, a thorough literature search was carried out in order to identify as many relevant investigations on the subject as possible. Every kind of imaging sensor was considered (visible range, multispectral, hyperspectral, chlorophyll fluorescence, etc.), provided that images were captured at close range, thus excluding research using Unmanned Aerial Vehicles (UAVs), airplanes and satellites. A careful analysis of the techniques for detection and classification was carried out and used as basis for an in-depth discussion on the main challenges yet to be overcome. Some directions for future research are also suggested, having as target to increase the practical adoption of this kind of technology. [ABSTRACT FROM AUTHOR]

• Red color level of image background at wheat booting can simulate relative yield. • Images were taken from heterogeneous stands of cultivar mixtures with 6 replicates. • The new index (BRL) was more efficient compared with some other vegetation criteria. • The similarities between patterns of RGY & BRL was acceptable even in the mixtures. • It is suggested that the non-vegetation image segments may provide valuable data. Digital cameras are increasingly utilized as a reliable and readily available remote sensing tool in the emerging technology of high throughput phenotyping and agronomic field assessments. In almost every image analysis of green canopies, image segmentation into the vegetation and non-vegetation parts using thresholding is a fundamental step. Therefore, only the vegetation parts of the high resolution images are considered for estimations and the data recorded in the remained pixels will be ignored. In the present study, an accidental observation is described, in which the red color level of the non-vegetation segments of the images taken at the wheat booting phase could simulate the relative pattern of final grain yield. An image archive of heterogeneous wheat canopies (i.e. cultivar mixtures), captured from experimental plots of a field study conducted at the School of Agriculture, Shiraz University, Iran, was analyzed. It was observed that the red color levels of the image background (BRL, mostly composed of soil pixels) were accurately simulating the final grain yields of the treatments, (i.e. very precise mimicry of relative grain yields –RGY– in monocultures, and exact determination of curve extrema and also the direction of fluctuations, in the case of mixtures). Compared with other colors and image-derived criteria evaluated, BRLs had the pattern most similar to the relative grain yields. Moreover, assessments showed the importance of using higher number of replicates in analyses (i.e. 6 vs 3 replications). This observation suggests the red color level of the image-background as a potential novel index for the prediction of wheat grain yield during the preanthesis period; and also the ability of the non-vegetation image segments to provide valuable data for crop remote sensing and image processing studies. [ABSTRACT FROM AUTHOR]

• Cost-effective monocular visual odometry system for tracked vehicles on agricultural terrains. • Simplified hardware and a low complexity system, without compromising performance. • Enhanced image processing algorithm with sub-pixel capability. In precision agriculture, innovative cost-effective technologies and new improved solutions, aimed at making operations and processes more reliable, robust and economically viable, are still needed. In this context, robotics and automation play a crucial role, with particular reference to unmanned vehicles for crop monitoring and site-specific operations. However, unstructured and irregular working environments, such as agricultural scenarios, require specific solutions regarding positioning and motion control of autonomous vehicles. In this paper, a reliable and cost-effective monocular visual odometry system, properly calibrated for the localisation and navigation of tracked vehicles on agricultural terrains, is presented. The main contribution of this work is the design and implementation of an enhanced image processing algorithm, based on the cross-correlation approach. It was specifically developed to use a simplified hardware and a low complexity mechanical system, without compromising performance. By providing sub-pixel results, the presented algorithm allows to exploit low-resolution images, thus obtaining high accuracy in motion estimation with short computing time. The results, in terms of odometry accuracy and processing time, achieved during the in-field experimentation campaign on several terrains proved the effectiveness of the proposed method and its fitness for automatic control solutions in precision agriculture applications. [ABSTRACT FROM AUTHOR]

• An image processing system for agricultural spreader calibration was developed. • This automatic tool uses a specific mat that simulates calibration standards rules. • Tests were performed with several image acquisition conditions and 2 types of granules. • The average accuracy was 0.75 ± 0.75% for fertilizer and 2.12 ± 1.68% for wheat. • The accuracy and robustness justify its use to support spreader pattern calibration. An Automatic Calibration of Fertilizers (ACFert) system was developed, for use with centrifugal, pendulum or other types of broadcast spreaders which distribute dry granular agricultural materials on the top of the soil. The ACfert is based on image processing techniques and includes a specially designed mat, which should be placed in the ground for spreaders calibration. A set of images acquired outdoor by a standard device (simple camera) is used to extract information about the spreader distribution pattern. Each image is processed independently, providing as output two numerical values for each grid element present in the image – the number of fertilizers/seeds counted, and its numerical label. The performance of ACFert was evaluated for automatic granules detection using a set of manual counting measurements of nitrate fertilizer and wheat seeds. A total of 185 images acquired with two mobiles devices were used with a total of 498 quadrilateral elements observed and analysed. The overall mean absolute relative error between counting and computed by the ACFert system, were 0.75 ± 0.75% for fertilizer and 2.12 ± 1.68% for wheat. This near real-time calibration tool is a very low cost system that can be easily used on field, providing results to support accurate spreader calibration in near real time for different types of fertilizers or seeds. [ABSTRACT FROM AUTHOR]

• Automatic Plant disease diagnosis for early disease symptoms. • Novel image processing algorithm in combination with machine learning methods. • A testing on wheat was performed with 7 mobile devices over a period of 4 seasons. • Obtained results reveal AuC metrics higher than 0.80 for all the analyzed diseases. Fungal infection represents up to 50 % of yield losses, making it necessary to apply effective and cost efficient fungicide treatments, whose efficacy depends on infestation type, situation and time. In these cases, a correct and early identification of the specific infection is mandatory to minimize yield losses and increase the efficacy and efficiency of the treatments. Over the last years, a number of image analysis-based methodologies have been proposed for automatic image disease identification. Among these methods, the use of Deep Convolutional Neural Networks (CNNs) has proven tremendously successful for different visual classification tasks. In this work we extend previous work by Johannes et al. (2017) with an adapted Deep Residual Neural Network-based algorithm to deal with the detection of multiple plant diseases in real acquisition conditions where different adaptions for early disease detection have been proposed. This work analyses the performance of early identification of three relevant European endemic wheat diseases: Septoria (Septoria triciti), Tan Spot (Drechslera triciti-repentis) and Rust (Puccinia striiformis & Puccinia recondita). The analysis was done using different mobile devices, and more than 8178 images were captured in two pilot sites in Spain and Germany during 2014 , 2015 and 2016. Obtained results reveal an overall improvement of the balanced accuracy from 0.78 (Johannes et al., 2017) up to 0.87 under exhaustive testing, and balanced accuracies greater than 0.96 on a pilot test performed in Germany. [ABSTRACT FROM AUTHOR]

Highlights • Developing an image processing algorithm for segmenting and separating the center potato slice from other slices. • Developing an ANN and a linear models between shrinkage and moisture content. • Developing an microwave power adjusting algorithm based on moisture content during drying process. Abstract In this study, machine vision was used for measuring area shrinkage of potato slices during thin layer drying process and then an artificial neural network (ANN) and linear models were investigated to predict moisture content (MC) of potato slices based on area shrinkage. Then an algorithm for adjusting the microwave power with respect to the predicted MC during the drying process was developed. A drying setup including imaging unit, lightning unit, infrared temperature sensor, image processing algorithm and microwave power adjusting program based on MC was developed. The experiments with two microwave power modes (variable and constant) have been done. The developed image processing had ability to separate connected potato samples and measured the shrinkage of center sample. The consequences expressed that the ANN with 1-3-1 structure had better results than linear model and could predict the MC based on shrinkage with 0.0966 RMSE and 96.87 R values on test data set. Also evaluating the developed ANN model with new experiments data set revealed that it could predicted the MC with 0.094 RMSE and 96 R values and resulted it has great accuracy and reliability. The real-time evaluating the image processing algorithm and ANN model with another new experiments indicated that the developed method has good promising ability for adjusting the microwave power and preventing the increased of microwave power density during the potato chips drying process. [ABSTRACT FROM AUTHOR]

Highlights • The feature selection algorithm was performed using Linguistic Hedges. • Characteristics of H S , curvature and angle of stem, H L , B L and L L were selected. • Plant conditions were evaluated using Adaptive neuro-fuzzy classifier. • The accuracy of the classification was 84.04. Abstract Agriculture is one of the largest and most important water consuming sector in the world. More than 80% of water resources are wasted in this sector because of the lack of advanced technology. Therefore, by saving water and controlling the growth conditions of the plant simultaneously, a low level of water consumption can be achieved. In this research, an automated machine vision system was designed and developed to determine the water requirement of the Lilium plant. Therefore, the average values of each RGB, Lab and HSV channel of leaves and stem along with curvature, length and orientation of stem were extracted. After that, features were analyzed using Duncan's multiple range test and the linguistic hedges feature selection algorithm combined with adaptive neuro-fuzzy classifier. According to the results, H s , curvature, stem angle, H L , B L and L L were the most significant features and the classifier accuracy for four plant conditions (100% of field capacity (FC), 80% of FC, 60% of FC and 40% of FC) were 84.84, 84.21, 86.95 and 80%, respectively. The results showed that the proposed system has the ability to determine the levels of water stress and provide the amount of water requirement. [ABSTRACT FROM AUTHOR]

Highlights • Image processing effectively predicted Kernel Processing Score (KPS). • Algorithm and sieving results were well correlated r(23) = 0.8, p < 0.001. • Mechanical sieving reduced particle size, potentially biasing lab results. • KPS assessment efficiency will encourage adjustment of processors during harvest. • Kernel processor adjustment will improve feed quality and increase milk production. Abstract An image processing algorithm was developed to characterize the size distribution of corn kernel particles from Whole Plant Corn Silage (WPCS). The algorithm determines particle cross sectional area and maximum inscribed circle diameter as well as cumulative undersize percent, dimensions of significance, and key characteristics of the distribution including mean particle size, skewness, and kurtosis. Kernel Possessing Score (KPS) was derived from the area-weighted cumulative undersize percent of 4.75 mm. Samples of WPCS harvested using self-propelled forage harvester with crop processing roll gap clearance of 1, 2, 3, and 4 mm were analyzed in their fresh, dry, and sieved states. Algorithm results were compared with the standard method of mechanical sieving and found to be well correlated r(23) = 0.8, p < 0.001. Additionally, analysis of particles before and after sieving indicated that sieving significantly increased KPS for 1 (t(5) = 6.6, p = 0.001), 2 (t(5) = 4.2, p = 0.01), and 3 (t(5) = 3.5, p = 0.02) mm samples. This algorithm has the potential to more accurately determine KPS in field compared to currently available methods, allowing for adjustment of kernel processing during harvest which will improve silage quality. [ABSTRACT FROM AUTHOR]

Highlights • The three waterlogging stress levels of rapeseed were investigated. • Hyperspectral RGB images and VNIR spectra (400–1000 nm) of waterlogging rapeseed leaves were extracted and analyzed. • QDA, KNN and SVM classifiers were applied and compared in classify analysis. • Selection of optimal wavebands for detection of waterlogging stress using SPA algorithm. Abstract The objective of this study was to investigate the utility of hyperspectral images (HSIs) for the detection of oilseed rape waterlogging stress. We assessed two oilseed rape varieties, the non-hybrid NingYou 22 (NY 22) and hybrid NingZa 19 (NZ 19), and HSIs of oilseed rape leaves under different durations of waterlogging stress (0, 3, and 6 days) were collected to build three datasets (NY 22, NZ 19, and both combined). We extracted red–green–blue (RGB) images and visible and near-infrared (VNIR 400–1000 nm) spectra from a region of interest (ROI) in each HSI. Quadratic discriminant analysis (QDA), k-nearest neighbor (KNN), and support vector machine (SVM) classifiers were used to build classification models for comparing images and spectra of samples under different waterlogging levels among the three datasets, and to conduct training and prediction. From each dataset, 70% of the images were used for training, and the remaining 30% were used for testing. In the classification of full-wavelength HSIs (400–1000 nm), QDA and SVM exhibited high multivariate classification accuracy, reaching 77.37% and 95.90% accuracy, respectively. In contrast, KNN displayed low accuracy, but good identification and prediction ability for variety NZ 19. Six optimal wavebands of 529, 641, 698, 749, 856, and 979 nm were used as input for successive projections algorithm (SPA) classification and analysis. The QDA mode had better classification performance, with identification accuracies of 100% and 94.44%, respectively. Overall, the VNIR classification results exceeded those of image classification. These results show that hyperspectral imaging technology is feasible and useful for the detection of oilseed rape waterlogging stress. [ABSTRACT FROM AUTHOR]

Highlights • Recent advances of weed detection using ground-based machine vision were reviewed. • Features for segmenting vegetation and detecting weeds from crops were elaborated. • Recently developed deep learning-based approaches for weed detection were covered. • Challenges and solutions for weed detection in the field were presented. Abstract Weeds are among the major factors that could harm crop yield. With the advances in electronic and information technologies, machine vision combined with image processing techniques has become a promising tool for precise real-time weed and crop detection in the field, providing valuable sensing information for site-specific weed management. This review summarized the advances of weed detection using ground-based machine vision and image processing techniques. Concretely, the four procedures, i.e., pre-processing, segmentation, feature extraction and classification, for weed detection were presented in detail. To separate vegetation from background, different color indices and classification approaches like color index-based, threshold-based and learning-based ones, were developed. The difficulty of weed detection lies in discriminating between crops and weeds that often have similar properties. Generally, four categories of features, i.e., biological morphology, spectral features, visual textures and spatial contexts, were used for the task, which were discussed in this review. Application of conventional machine learning-based and recently developed deep learning-based approaches for weed detection were also presented. Finally, challenges and solutions provided by researchers for weed detection in the field, including occlusion and overlap of leaves, varying lighting conditions and different growth stages, were discussed. [ABSTRACT FROM AUTHOR]

Highlights • We developed and used a custom-built low-resolution thermal-RGB imager to monitor cherry fruit skin temperature. • We developed an algorithm to calculate fruit surface temperature from thermal and RGB images. • We successfully linked fruit skin temperatures before and after simulated rainfall to surface wetness. • The in-field imaging system was successfully used to detect cherry fruit surface wetness. Abstract Sweet cherry fruit cracking caused by seasonal rains is a major source of crop loss in the U.S. Pacific Northwest region and around the globe. In-field monitoring of cherry fruit surface wetness and temperature is, therefore, very important in fruit loss management. To determine the feasibility of low-resolution thermal-RGB imagery for detecting sweet cherry surface wetness, an experiment was carried out in plots of Skeena and Selah cherry varieties with Y-trellised and vertical architecture, respectively, at the Roza Farm of Washington State University, Prosser, WA. To wet cherries, 5 mm of rain was applied by running a rain simulator for 4 min (1.25 mm min−1) above canopies. Rainwater samples were collected using five rain gauges to quantify the applied amount of water. The in-field sensing setup included two custom-built thermal-RGB imagers, a microclimate-measuring unit and two leaf wetness sensors. The imagers were installed at a height of 2.1 m above the ground surface and were about 20 cm from the target cherries. The leaf wetness sensors were next to the cherries in the field of view of the imagers. A custom computer vision algorithm was developed and used to identify leaves and cherries in thermal and RGB images and extract the surface temperatures. The relationship between raw and normalized surface temperature, and wetness level and duration was investigated. The applicability of normalized cherry surface and air temperature difference was also studied. The results revealed that low-resolution thermal-RGB imagery can be used for detecting cherry fruit wetness level and duration. There was also a high correlation between the surface temperature of leaves and cherry fruits during the wetness period suggesting the temperature of leaves as reliable surrogate for cherry surface wetness and temperature monitoring. By utilizing the proposed imagery-based system, decision aid tools may be developed for efficient rainwater removal to prevent fruit cracking. [ABSTRACT FROM AUTHOR]

Highlights • The new algorithm is evaluated on Flavia dataset and on self-collected dataset of 625 leaf images. • A thorough review of different plant identification techniques is presented. • Different Classifier are compared for classification task. • Results show that the proposed algorithm can attain plant recognition accuracy of up to 98.75% with Flavia and 97.25% on self-collected dataset. Abstract A large number of studies have been performed during the past few years to automatically identify the plant type in a given image. Besides common object recognition difficulties arising mainly due to light, pose and orientation variations, the plant type identification problem is further complicated by the differences in leaf shape overage and changing leaf color under different weather conditions. The limited accuracy of existing approaches can be improved using an appropriate selection of representative leaf based features. This study evaluates different handcrafted visual leaf features, their extraction techniques, and classification methods. Towards this end, a new five-step algorithm is presented (comprising image pre-processing, segmentation, feature extraction, dimensionality reduction, and classification steps) for recognition of plant type through leaf images. The proposed algorithm is evaluated on a publicly available standard dataset 'Flavia' of 1600 leaf images and on a self-collected dataset of 625 leaf images. With the proposed algorithm, different classifiers such as k-nearest neighbor (KNN), decision tree, naïve Bayes, and multi-support vector machines (SVM) are tested. The best performing KNN, claimed for the final results, reveals that the proposed algorithm gives precision and recall values of 97.6% and 98.8% respectively when tested on 'Flavia' dataset. The proposed technique is also tested on our self-collected dataset, giving respectively 96.1% and 97.3% precision and recall measure results. Results confirm that our approach, when augmented with efficient segmentation techniques on raw leaf images, can be a significantly accurate plant type recognition method in practical situations. AlexNet, a Convolutional Neural Network (CNN) based approach is also compared for classification on the datasets as oppose to handcrafted feature-based approach and it is found that the later outperforms the former in robustness when the training dataset is small. [ABSTRACT FROM AUTHOR]

Highlights • Digital images of wheat canopy were analyzed to estimate Nitrogen and water status. • Multivariable models were built using PLSR and RF algorithms. • TPerformance of PLSR was remarkable for both single-date and combined-dates data. • RF showed better performance for combined-dates than single-date data. • The proposed method showed high potential to assess the crop status. Abstract In this research, winter wheat Nitrogen and water status assessment by a digital camera was studied. Field experiments were conducted during growing seasons, 2012, 2013 and 2014 in Potsdam, Germany. Different treatments of N fertilization and water regimes were applied to the plant. Aboveground biomass was sampled three times during each season, and analyzed for plant nitrogen content, fresh and dry biomasses, and water content. Color images of the plant canopy were taken in the field by a digital camera. The images were then processed and various features were derived. Multivariable models using the features and two different algorithms, Partial Least Square Regression (PLSR) and Random Forest (RF) were developed. The models were built for data of single sampling dates and combination of sampling dates of each season. In addition, spectral measurements were taken by a portable field spectroradiometer and a few vegetation indices (VIs) were calculated. Linear regression models were built between the VIs and the plant parameters. According to the results obtained, PLSR presented the strongest models for both single-date and combined-dates data. R2 (corresponding RMSE) of the models for N content and water content varied in the range of 0.77–0.91 (0.25–0.14), and 0.53–0.95 (0.75–0.36), respectively, for single-date data, and 0.76–0.89 (0.24–0.15), and 0.65–0.92 (1.49–1.66), respectively, for combined-dates data. Performance of RF was better for combined-dates than single-date data. The conclusion is that a combination of digital image processing with an appropriate machine learning method has a high potential for plant growth status assessment in the field. [ABSTRACT FROM AUTHOR]

Highlights • Scoring of lateral and sternal lying posture of pigs is presented. • Image processing and SVM were used to monitor lying postures of each pig. • Area and perimeter of boundaries and convex hull were used as inputs to the SVM. • High overall accuracies show the possibility of using the SVM. Abstract The behaviour of animals provides information on their health, welfare and environmental situation. In different climatic conditions, pigs adopt different lying postures; at higher temperatures they lie laterally on their side with their limbs extended, while in lower temperatures they will adopt a sternal or belly lying posture. Machine vision has been widely used in recent years to monitor individual and group pig behaviours. So, the aim of this study was to determine whether a two-dimensional imaging system could be used for lateral and sternal lying posture detection in grouped pigs under commercial farm conditions. An image processing algorithm with Support Vector Machine (SVM) classifier was applied in this work. Pigs were monitored by top view RGB cameras and animals were extracted from their background using a background subtracting method. Based on the binary image properties, the boundaries and convex hull of each animal were found. In order to determine their lying posture, the area and perimeter of each boundary and convex hull were calculated in lateral and sternal lying postures as inputs for training of a linear SVM classifier. The trained SVM was then used to detect the target postures in binary images. By means of the image features and the classification technique, it was possible to automatically score the lateral and sternal lying posture in grouped pigs under commercial farm conditions with high accuracy of 94.4% for the classification and 94% for the scoring (detection) phases using two-dimensional images. [ABSTRACT FROM AUTHOR]

Highlights • Methods and equipment (MMS) used in a mobile measuring system were introduced. • The image processing procedures and calculation formula for body components measurements were specified. • Body components were estimated by MMS in large-scale pig farms with high accuracy. Abstract New equipment applications for the monitoring and measuring of pig growth play an important role in quality management and productivity improvement. One challenge that researchers face is how to obtain the body components of animals in large-scale farms automatically, which are significant to their growth, feed conversion, and disease occurrence. This study provided a mobile measuring system that could acquire the real-time data in large-scale farm to measure the body components of 200 pigs in 70, 100, and 130 days old respectively. The body components are: body length, body width, body height, hip width, and hip height. The programs of system were developed based on LabVIEW development platform, which consisted of three parts: control, calibration, and image processing. The calibration part was utilized to identify a distortion coefficient and calculate camera parameters. Backside images of pigs during growth were taken by using binocular cameras after calibration, and then processed to estimate pig body components. Experimental results showed that the mobile measuring system could be employed in large-scale pig farms and used to measure body components ranging from 23.1 to 116.2 cm. The estimated and manually measured body components suggested a good linear relationship with R2 > 0.82 and average deviation < 2.83%. These findings indicated that the proposed mobile measuring system could be utilized to observe pig growth and estimate body components. Therefore, the mobile measuring system offers the hardware and software for the non-contact and automatic measurement of pig growth data in large-scale farms. [ABSTRACT FROM AUTHOR]

Highlights • A new algorithm to extract piglet ear base temperature by using thermal image processing. • Identification of piglet head part by using SVM based on contour features. • Ear base points are located based on the shape features of piglet head part. • More than 97% of the testing images whose ear base temperatures can be obtained with the error less than 0.4 °C. Abstract Ear bases are considered the thermal windows of a piglet. Temperature variation in piglet ear bases can be used as the indicator of a piglet's health status. However, piglet skin temperatures in thermal windows in the existing research are obtained manually from infrared thermal images captured by a thermography. This has put an obstacle at the automatic identification of piglets with health disorder. An algorithm was proposed in this paper to extract ear base temperature automatically from top view piglet thermal images. Firstly, a SVM (Support Vector Machine) classifier was trained to identify piglet head part. Then, two ear base points were located based on the shape feature of the head part contour. Finally, two maximum temperatures inside the two circles centered by ear base points were extracted as the ear base temperatures. The proposed algorithm was implemented in Matlab® (R2016a) and applied to 100 testing images. The extracted ear base temperatures were compared with those extracted manually by using Fluke SmartView 3.14 (FLUKE Systems). Comparison results showed that for left and right ear base respectively, 97% and 98% of the testing images had an error within 0.4 °C. Ear base temperatures with such accuracy provided a foundation for the automatic identification of sick piglets. [ABSTRACT FROM AUTHOR]

Highlights • Convolutional neural network for Esca disease detection in vineyard. • Unmanned aerial vehicle and RGB sensor as imaging system. • Deep learning model built on combination between color spaces and vegetation indices. • Comparative study between different parameters, filters, block size, color spaces, vegetation indices.... Abstract Detection of symptoms in grape leaves is a very important factor in preventing a serious disease. An epidemic spread in vineyards has huge economic consequences and therefore it is considered a major challenge for viticulture. Automatic detection of vine diseases can play an important role in addressing the issue of diseases management. This study deals with the problem of identifying infected areas of grapevines using Unmanned Aerial Vehicles (UAV) images in the visible domain. In this paper we propose a method based on Convolutional neural network (CNN) and color information to detect symptoms in the vine yards. We studied and compared performances of CNNs using different color spaces, vegetation indices, as well as the combination of both information. The obtained results showed that CNNs with YUV color space combined with ExGR vegetation index, and CNNs with a combination of ExG, ExR, ExGR vegetation indices yield the best results with accuracy more than 95.8 %. [ABSTRACT FROM AUTHOR]

Highlights • An image processing approach was designed for olive cultivars identification. • The morphological features handled by the specialist were extracted by image processing. • Most discriminant features were identified. • The proposal was assessed with different classifier designs. Abstract The identification of olive cultivars is of utmost importance for a multitude of factors affecting both, the olive oil elaboration process and fair trade exchanges. The accurate varietal identification is a time consuming task that requires trained specialists or expensive and specific equipment. When applying the traditional method, a specialist assesses morphological features using the olive endocarp. A proposal to automate this identification method is presented in this paper. Endocarp images, acquired under three different perspectives, are processed to extract the same information that the specialist utilizes. Then, the partial least squares discriminant analysis classifier, with or without feature selection, has been tested on a set of 250 samples from 5 different varieties. Results show that the proposal is an alternative identification method which could also be used in the traditional one in order to assist the specialist in the determination of the variety. [ABSTRACT FROM AUTHOR]

• This study proposes M-CR U-Net architecture, with image post-processing, to enhance early symptom detection of BSR in crowded regions. • The proposed M-CR U-Net provides excellent segmentation of finer details with minimal overlapping regions. • The proposed OVCS algorithm effectively separates overlapped contours. • The proposed method detects early BSR symptoms in crowded oil palm regions, improving plantation management. • The improved detection of BSR disease can help maintain the health of oil palm trees and ultimately increase yield and growers' income. The oil palm tree (Elaeis guineensis Jacq.) is a vital and economically important plant, as the oil from its fruit is a primary export for many Southeast Asian countries. Hence the health of these trees is directly linked to their yield and the income of the growers. Unfortunately, basal stem rot (BSR) disease has been identified as the main threat to oil palm trees, causing infected trees to reduce their production and ultimately leading to death. While prior research has shown high accuracy in detecting oil palm trees, there has been limited focus on automating the observation of individual tree health status, particularly in identifying early symptoms of BSR disease within densely populated regions. This study introduces an enhanced approach for early symptom detection of BSR disease in densely populated oil palm tree areas. We present a modified U-Net architecture called the Multi-Convolution Residual (M-CR) U-Net , coupled with an image postprocessing method known as the overlapped contour separation (OVCS) algorithm. The M-CR U-Net differs from the original U-Net as it leverages on multiple convolution kernels of various sizes while incorporating skip connections. This design improvement results in superior segmentation performance compared to approaches utilizing fixed kernel sizes. Our proposed M - CR U-Net outperforms alternative architectures across a range of metrics, demonstrating improved training efficiency and delivering enhanced quantitative and qualitative segmentation outcomes. These achievements underscore its proficiency in accurately delineating the intricate features within input images. The M - CR U-Net excels in capturing the fine details while minimizing overlapping regions, as evidenced by its dice coefficient of 0.978, intersection over union of 0.871, precision of 0.989, and recall of 0.982. Additionally, the OVCS algorithm enables effective segmentation of individual oil palm trees within highly crowded regions. By adeptly separating the boundaries of overlapped areas, this algorithm enhances segmentation results. This innovation ensures precise identification of the health and boundaries of oil palm trees, leading to an average performance improvement of 2.716% across various evaluation metrics. The proposed methodology demonstrates significant potential in detecting early symptoms of BSR disease in oil palm trees, even within crowded regions, leading to more effective management of oil palm plantations. [ABSTRACT FROM AUTHOR]

• Multimodal hierarchical fusion method based on attention mechanism is proposed. • Improved recognition performance of tomato stem against similar-colored background. • Proposed method improves recognition by fusing RGB and NIR image features. • Multimodal loss function reduces the influence of information mutual exclusion. • Local depth separable convolution and Conv-BN layer merging reduce inference time. The tomato plant's main-stem is a feasible lead for robotic searching the grows discretely-growing targets of harvesting, pruning or pollinating. Owing to the highlighted reflection characteristics of the main-stem in the near-infrared (NIR) waveband, this study proposes a multimodal hierarchical fusion method (YOLACTFusion) based on the attention mechanism, to achieve an instance segmentation of the main-stem from similar-colored differentiation (i.e., green leaf and green fruit) in robotic vision systems. The model inputs RGB images and 900–1100 nm NIR images into two ResNet50 backbone networks and uses a parallel attention mechanism to fuse feature maps of various scales together into the head network, to improve the segmentation performance of the main-stem of RGB images. The loss function for the multimodal image weights the original loss on the RGB image and the position offset loss and classification loss on the NIR image. Furthermore, the local depthwise separable convolution is used for the backbone network, and Conv-BN layers are merged to reduce the computational complexity. The results show that the precision and recall of YOLACTFusion of the main-stem detection, respectively reached 93.90 % and 62.60 %; and the precision and recall of instance segmentation reached 95.12 % and 63.41 %, respectively. Compared to YOLACT, the mean average precision (mAP) of YOLACTFusion is increased from 39.20 % to 46.29 %, the model size is reduced from 199.03 MB to 165.52 MB , while the image processing efficiency remains similar. The overall results show that the multimodal instance segmentation method proposed in this study significantly improves the detection and segmentation of tomato main-stems under a similar-colored background, which would be a potential method for improving agricultural robot's visual perception. [ABSTRACT FROM AUTHOR]

• Extensive review has been done related to the perception system utilized in autonomous agriculture robots. • The study will help the researchers know about the recent trends in agricultural robots. Agriculture represents an essential aspect of human existence, providing the sustenance necessary for survival in the form of food and various other products. Additionally, it serves as a foundational pillar of economic development, offering employment and income opportunities to countless individuals. The incorporation of machine vision in agriculture has emerged as a crucial technology, enabling farmers to automate various tasks, such as crop monitoring and yield prediction using cameras and image processing techniques and even providing real-time information on crop maturity for harvest planning. This not only helps increase efficiency and productivity but also provides valuable insights for precision agriculture, enabling more quick and informed decision-making and ultimately leading to improved crop yields and financial returns. This scholarly work conducts a thorough examination of the various components that comprise a machine vision system, specifically delving into the techniques of image acquisition, processing, and classification. It also explores the methods employed within each of these techniques, and how the combination of such processes is used to perform various agricultural activities such as weeding, seeding, harvesting, fruit counting, overlapping, sorting, etc. Furthermore, it aims to guide on how the knowledge gained can be applied to build practical machine vision systems for agriculture. Additionally, this work highlights the research gaps, address current problems that can be solved and its potential for future advancement in the field of agriculture, and guides readers towards promising areas for future endeavours. [ABSTRACT FROM AUTHOR]

• An efficient lightweight model based on structural reparameterization for apple leaf disease identification (RepDI). • RepDI is a CPU-friendly model by the using of structural reparameterization. • A lightweight parallel channel & spatial attention module is proposed. • A novel apple leaf disease dataset is collected and released. Apple disease is one of the major factors affecting apple production, and the visual diagnosis of apple leaves is an efficient disease identification solution. In this paper, we propose an efficient lightweight model based on structural reparameterization for apple leaf disease identification, called RepDI for short. To achieve faster inference on the CPU devices, we introduce depth-wise separable convolution and structural reparameterization technology in RepDI, which has different structures during training and inference. In addition, to better capture diseased leaves and disease regions in complex contexts, we propose the parallel dilated attention mechanism module and embed it into RepDI. Experiments show that RepDI can achieve state-of-the-art performance in disease identification task, compared to most lightweight models. Meanwhile, RepDI achieves the fastest inference speed on our desktop CPU, which is an important factor in practical applications. Furthermore, we collect and annotate a novel dataset for apple leaf diseases from real scenarios, called Real-ALD, which is more challenging than previous datasets. And RepDI achieves a top-1 accuracy of 98.92 in the Real-ALD dataset under a limited training configuration. Our code is released to contribute to the plant protection community and we will further explore the potential of RepDI for down-stream detection, segmentation tasks. [ABSTRACT FROM AUTHOR]

• An automatic detection system for fertilizer deposition distribution was designed. • Combining Swin Transformer and PANet network were used in the detection. • A granular fertilizer mask generation network is proposed for small and dense fertilizer. • Fertilizer quality can be obtained using the segmentation detection network, • Tests showed that the F-TPP network model improved the segmentation precision of granular fertilizer. Beginning with the problems of high cost, low efficiency and high working intensity of traditional granular fertilizer deposition distribution detection methods, this study proposes an automatic detection system of granular fertilizer deposition distribution pattern based on deep learning method. The system analyzes the distribution pattern of granular fertilizer deposition on the ground and allows for parameter calibration, performance evaluation and structural optimization of fertilizer spreading devices. In this study, hardware and software parts of the proposed system were designed. The hardware part primarily includes a customized fertilizer calibration mat, an image acquisition device, a system control terminal, and a processing core to realize image acquisition, storage, data processing, and system control functions. For the software part, a granular fertilizer identification algorithm was designed; in addition, the granular fertilizer instance segmentation network model F-TPP (Fertilizer-Swin Transformer PANet partition) was proposed that integrated the Swin Transformer network and path aggregation network (PANet) to address the problems associated with extremely small and relatively dense targets in the detection of granular fertilizers. Moreover, the F-TPP network model optimized the Anchor generation scale for appropriate segmentation detection of granular fertilizers. The experimental results show that the F-TPP network model outperformed comparison conventional network models, such as Mask R-CNN, SOLO, and YOLOv5. Indeed, the detection efficiency of the proposed model was higher than that of conventional models for small and dense granular fertilizers. The proposed model yielded a detection precision, recall rate, and F1 score of 93.8%, 96.3%, and 0.95, respectively. The F-TPP network model improves the segmentation precision of granular fertilizer; thus, it improves the accuracy of automatically-deposited fertilizers. According to the experimental results, the proposed detection method yielded maximum absolute and relative errors of 0.59 g and 5.25%, respectively. [ABSTRACT FROM AUTHOR]

• Strain is a new monitoring quantity that can be applied to seeding monitoring operations. • Combined bionic design improves the omnidirectional sensitivity of strain sensors. • The combined bionic sensors enable the precise monitoring of strains induced by seeding. • The strain peak recognition algorithm can accurately judge the effect of each seeding. • The strain-seeding monitoring system developed increases monitoring accuracy by 3.51%. This study has discovered a novel monitoring quantity suitable for seeding spacing monitoring, which is the strain signal generated by the seed flow through the seed discharge port during the operation of the finger-clip seed metering device. The strain sensor output a dataset of approximately 30 electrical signals for each flow of the seed pick-up clip of the finger-clip seed metering device through the seed discharge port. The number of peak points in this data set is equivalent to the number of seeds sown, enabling accurate judgment of missed seeding or reseeding. This study addressed the problem that field vibrations tend to cause displacement of seed, which in turn triggers a change in strain direction. Through a combined bionic design, a scorpion body surface V-shaped crack structure (ultrasensitive sensing function) and a spider-web annular topology (omnidirectional equalization sensing function) were selected as bionic prototypes to develop a combined bionic strain sensor (CBSS). The test results showed that the CBSS combines the functional characteristics of both bionic structures and was more suitable for seed spacing monitoring. Compared to the conventional straight-line V-shaped crack strain sensor (CSS), CBSS improved the omnidirectional sensitivity by 21.54–35.57%. To mitigate the impact of ground frequency noise in the system causing fluctuations in the CBSS output signal, a strain peak recognition algorithm (SPRA) was developed using image processing techniques and Python. The SPRA effectively removed ground frequency noise using a gray-scale erosion algorithm and accurately recognized peak points in the electrical signal dataset through the Suzuki and CEV algorithms. Results showed that the SPRA reduced monitoring errors by 20.52% compared to the conventional direct data processing method (DDPM). Based on the CBSS and SPRA, a strain seeding spacing monitoring system (S-SMS) was developed. Bench tests revealed that the parameters of the combined bionic annular crack structure (crack depth h) of the CBSS significantly affected the monitoring error of the S-SMS, with optimal performance achieved at the h value of 82 μm. Field tests demonstrated that the CBSS accurately monitored the strain signal triggered by each seed with 98.51% accuracy, while the SPRA recognized missed seeding and reseeding with 98.64% and 98.42% accuracy, respectively. Compared to the conventional photoelectric diffuse reflection seeding monitoring system (P-SMS), the S-SMS reduced seeding monitoring errors in the field by 3.51%. This research can provide a new research path for the development of seeding spacing monitoring systems. [ABSTRACT FROM AUTHOR]

Nowadays, most agricultural robots rely on precise and expensive localisation, typically based on global navigation satellite systems (GNSS) and real-time kinematic (RTK) receivers. Unfortunately, the precision of GNSS localisation significantly decreases in environments where the signal paths between the receiver and the satellites are obstructed. This precision hampers deployments of these robots in, e.g., polytunnels or forests. An attractive alternative to GNSS is vision-based localisation and navigation. However, perceptual aliasing and landmark deficiency, typical for agricultural environments, cause traditional image processing techniques, such as feature matching, to fail. We propose an approach for an affordable pure vision-based navigation system which is not only robust to perceptual aliasing, but it actually exploits the repetitiveness of agricultural environments. Our system extends the classic concept of visual teach and repeat to visual teach and generalise (VTAG). Our teach and generalise method uses a deep learning-based image registration pipeline to register similar images through meaningful generalised representations obtained from different but similar areas. The proposed system uses only a low-cost uncalibrated monocular camera and the robot's wheel odometry to produce heading corrections to traverse crop rows in polytunnels safely. We evaluate this method at our test farm and at a commercial farm on three different robotic platforms where an operator teaches only a single crop row. With all platforms, the method successfully navigates the majority of rows with most interventions required at the end of the rows, where the camera no longer has a view of any repeating landmarks such as poles, crop row tables or rows which have visually different features to that of the taught row. For one robot which was taught one row 25 m long our approach autonomously navigated the robot a total distance of over 3.5 km, reaching a teach-generalisation gain of 140. • We extend the classic visual teach & repeat to a novel teach & generalise approach. • Our approach is capable of extracting unique features using a deep learning method. • The approach only uses an uncalibrated monocular camera and wheel odometry. • Successfully traverses several unseen paths having only been taught one similar path. • Our method is entirely developed in ROS enabling fast plug & play implementation. [ABSTRACT FROM AUTHOR]

Highlights • A CNN was trained for identification of plant diseases. • Tests considered 12 plant species and 56 diseases. • The effects of using limited datasets were investigated. • The effects of background removal were investigated. • Accuracies varied from 60% to 100% depending on the characteristics of each crop. Abstract The problem of automatic recognition of plant diseases has been historically based on conventional machine learning techniques such as Support Vector Machines, Multilayer Perceptron Neural Networks and Decision Trees. However, the prevailing approach has shifted to the application of deep learning concepts, with focus on Convolutional Neural Networks (CNNs). In general, this kind of technique requires large datasets containing a wide variety of conditions to work properly. This is an important limitation, given the many challenges involved in the construction of a suitable image database. In this context, this study investigates how the size and variety of the datasets impact the effectiveness of deep learning techniques applied to plant pathology. This investigation was based on an image database containing 12 plant species, each presenting very different characteristics in terms of number of samples, number of diseases and variety of conditions. Experimental results indicate that while the technical constraints linked to automatic plant disease classification have been largely overcome, the use of limited image datasets for training brings many undesirable consequences that still prevent the effective dissemination of this type of technology. [ABSTRACT FROM AUTHOR]

Highlights • AWC is the most important disease in this crops and present high relation with climatic variables. • Electronic devices were designed for quantification of climatic variables in avocado production systems. • An app was developed with three functions: (ii) early warning system for AWC, (ii) diagnosis of AWC, and (iii) management practices of AWC. Abstract Quantification of climatic variables such as precipitation and temperature and determination of their interactions inside the soil profile is very important in avocado systems as a tool for prompt and appropriate management of several aspects of crop production such as the avocado wilt complex (AWC). Commercial electronic devices are very expensive, with close code, difficult to integrate to an open code system and present the disadvantage that are made and calibrated for different conditions and usually, most epidemiological models are not suitable for the Colombian conditions. The aim of this work was to design a low cost electronic device for the collection and transmission of climatic variables including moisture and temperature from inside the soil profile. As a complement, a mobile application was designed with three functions. (i) Available information in real time, which may be used as an early warning system for the AWC. (ii) Diagnosis of causal agents based on patterns associated with image processing, (a) decomposition of RGB color and comparison of images of plants diseased based on a machine learning algorithm, and (b) a key for identification of diseased plants based on different symptoms, was incorporated. (iii) Management of AWC. We evaluated the capacity of the system to quantify the values of different environmental variables, prediction of early warning system of AWC, and diagnosed strategy for management of AWC, based on a correlation model with the real data taken in avocado fields. The low cost electronic device showed a correlation higher than 90% compared with data obtained from traditional climatic stations and standard methods of soil analysis. In addition, the early warning system achieved a prediction higher than 70% of the variables associated with AWC, whose values increased when the prediction was corroborated with the diagnostic system. [ABSTRACT FROM AUTHOR]