Search

In current plant phenotyping research, the study of plant time-series images based on deep learning has received widespread attention. While such image data is relatively easy to obtain, the cost of annotation is high. One efficient method for achieving cost-effective training is through contrastive learning. Plant growth is slow, and the changes in image sequences over a period of time are small, with simple semantic information. Previous contrastive pre-training models struggled to effectively distinguish positive samples from the same image with different augmented views and similar negative samples from different images. Therefore, this paper proposes a method called self-supervised contrastive learning method for plant time-series images with a Priori Distance Embedding (PDE). The semantic information in images corresponding to different phenological stages of plants varies. This method transforms this crucial domain knowledge into prior distances for image pairs and conducts contrastive learning pre-training. The learned weights can be transferred to downstream tasks. Building upon this method, experiments were conducted on cherry time-series images to assess the quality of pre-training through a plant phenotyping image semantic segmentation task. To provide a comprehensive example of plant time-series image phenotypic analysis, this paper establishes a cherry growth temporal model, specifically including PDE pre-training, anomaly detection, semantic segmentation, and recording the results from the temporal dimension. The experiments indicate that this self-supervised contrastive learning method can be effectively applied to the pre-training of plant time-series images, demonstrating broad applicability in various computer vision studies related to plant phenotyping. • A pre-training method for plant images with a Priori Distance Embedding (PDE). • Transforms the phenological information into prior distances for image pairs. • An image acquisition device was built to obtain cherry time series images. • The PDE method was tested by plant phenotype image segmentation task. • We provided an exhaustive example of cherry time-series image phenotypic analysis. [ABSTRACT FROM AUTHOR]

• An efficient method is proposed for orchard tree trunk detection in natural scenes. • A multi-scenes orchard tree trunk dataset is established for experiments. • The YOLOv5 model structure is improved to enhance detection accuracy and speed. • Proposed method is robust against complex orchard environment. The recognition and localization of fruit tree trunks in orchard are important for orchard operation robots, which are the bases for automatic navigation, fruit tree spraying and fertilization etc. A method was proposed based on machine vision to detect target objects such as fruit tree trunks, person and supporters in orchard by improving the YOLOv5 deep learning algorithm in this paper, which is applicable to the recognition tasks in natural orchard environments. Firstly, 1354 images of the natural orchard collected by camera were image enhanced, weather effects such as rain, snow, bright light, shadow and fog were added to expand the dataset and to increase the robustness of the model. Secondly, the original YOLOv5 model was improved by replacing the Bottleneck network in the C3 module with the lightweight GhostNet V2 to reduce the network parameters, and changing the box loss function CIoU to SIoU in the loss function to make the regression of the detection box more accurate, and coordinate attention mechanism (CA) was added to the network to reduce the interference of useless background information in images. Before training, pre-anchor boxes were generated by using IoU-based K-means clustering, after that the dataset was fed into the improved YOLOv5 for training, and the trained model was used to detect the trunks. Finally, weighted boxes fusion (WBF) was used instead of the non-maximum suppression (NMS) in this paper for the output of the detection boxes. Then the density-based spatial clustering of applications with noise (DBSCAN) algorithm was used for trunk clustering. The improved target detection method was trained and validated on the experimental dataset. The model size is reduced by 43.6 %, model parameters are reduced by 46.9 %, and the mAP reaches 97.1 %, with an average detection speed of 198.2 ms per image. Compared with the original YOLOv5, the model is more lightweight, the detection accuracy and speed are improved. The improved YOLOv5 is also better than YOLOv3, NanoDet and SSD in terms of combined accuracy and speed, and has similar performance to YOLO_MobileNet in orchard dataset. The experimental results show that the improved YOLOv5 target detection model proposed in this paper is lightweight while still having better detection accuracy and detection speed in complex environments, and the model is small enough to be deployed to mobile or low-performance terminals for target detection in natural orchard environments. [ABSTRACT FROM AUTHOR]

• Precision agriculture method applicable to remote on-tree pomegranate pixel-based fruit size monitoring. • Comparison of Deep Learning and zero-shot vision Transformer instance segmentation models. • Image-Based size estimation algorithm for occluded fruits with protuberances. In situ automatic fruit monitoring is of great interest for more accurate and cost-efficient decision making in agriculture. For this purpose, the development of computer vision-based tools is essential. Deep Learning techniques have shown good performance in fruit detection and segmentation. Recently, new models based on Transformer architecture have emerged with promising potential and zero-shot inference capability. In this paper, a Deep Learning model, Mask R-CNN, was trained for on-tree pomegranate fruit segmentation and compared with foundational models based on Vision Transformer, Grounding DINO and Segment Anything Model. Results with Mask R-CNN proved a better performance, according to F1 score and AP metrics, and a lower computational cost, according to prediction time. One of the most interesting derived applications from fruit segmentation is fruit size estimation. However, segmented fruit masks are frequently incomplete due to occlusions. Therefore, image fruit size estimation is not a straightforward process. In this work, we also propose a novel algorithm to estimate and monitor the fruit size in pixel units from the automated masks. A median relative error of 1.39% was obtained, demonstrating the potential and feasibility of future fully-automatic fruit size estimators. [ABSTRACT FROM AUTHOR]

AgroCounters is an open-source repository for counting objects in images in the agricultural domain by utilizing deep-learning algorithms. In this paper, we present the framework of AgroCounters, which integrates state-of-the-art deep learning models, including regression-based counting, detection-based counting, and density-estimation-based counting, to accurately count various agricultural objects, such as fruits, vegetables, and livestock, in single images. The framework utilizes transfer learning techniques to optimize model performance on the limited labeled data available in the agricultural domain. We provide an open-source implementation of AgroCounters, which includes a multitude of algorithms for counting applications and a toolbox that includes metrics, training data tools, visualizations, and a simple installation guide for several open-source implementations of counting methods. We evaluated the performance of AgroCounters on multiple agricultural datasets acquired from RGB sensors, including plant leaves, melons, wheat grains, cherry tomatoes, grapes, apple flowers, bananas (fruit and leaves), pears, and chickens. We compared the results of the various implemented methods over these datasets and showcased the most suitable solution for each. YOLOv5, the most recent of the compared object detectors, provided the best results on all the examined datasets, and there was no clear 'winner' between Faster-RCNN and RetinaNet. Based on the analyzed datasets, when higher accuracy is required, the direct regression network (DRN) should be used; for small datasets, multiple scale regression (MSR) gives superior results. Based on the developments, we proposed guidelines for developing deep-learning-based counting solutions for agricultural applications, focusing on solutions and best practices for the agricultural domain. Overall, AgroCounters presents a promising solution for automated counting in the agricultural domain, offering significant potential for reducing manual labor, improving crop management, and increasing productivity. • AgroCounters is a repository for counting agriculture objects using deep learning. • AgroCounters includes regression, detection, and density estimation algorithms. • AgroCounters provides toolboxes for metric calculation, training, and visualization. • AgroCounters publicizes 10 different datasets with annotations and trained models. • Present guidelines for developing deep learning-based counters in agriculture. [ABSTRACT FROM AUTHOR]

• An adaptive cruise weeding robot for paddy fields was developed for the first time. • A real-time rice seedling recognition method based on MW-YOLOv5s was proposed. • A navigation line fitting method for paddy field adaptive operations was proposed. Weed control in paddy fields is a critical agronomic practice for enhancing rice yield, in which mechanical weed control is widely used due to its high weed control rate and convenient operation. However, traditional mechanical weed control methods require manual operation, leading to increased operational costs. Therefore, adaptive cruise weeding robots hold significant application promise and market potential. In this paper, an adaptive cruise weeding robot for paddy fields based on improved YOLOv5 is designed to adaptively perform weeding operations. Firstly, a real-time rice seedling recognition method based on MW-YOLOv5s is proposed by establishing a rice seedling dataset under different paddy field environments and growth stages. The model combines the Backbone network structure with MobileViTv3, replacing GIoU_loss with WIoU_loss, which significantly improves the precision and speed of rice seedling recognition. Next, the MW-YOLOv5s was integrated into a paddy weeding machine, and seedling navigation lines were extracted using the least square method. Finally, the control system autonomously operates the weeding machine in real time through feedback control based on the navigation path. The test results showed that the MW-YOLOv5s model exhibits strong recognition performance for rice seedlings under different paddy field backgrounds and growth stages, with a precision of 90.05 % and an mAP of 92.32 %, while the real-time performance reaches 19.51 FPS, which meet the requirements for real-time operation of the weeding machine in paddy fields. The results of the paddy field weed control experiment reveal a weed control rate of 82.4 % and a seedling injury rate of 2.8 %, which meet the agronomic requirements for mechanical weed control in paddy fields. The research findings contribute to the design of intelligent agricultural equipment and promote the practice and application of machine vision in the field of paddy field plant protection. [ABSTRACT FROM AUTHOR]

• Detailed review of over 300 research papers. • Object Detection and Tracking are suitable techniques to enhance Precision Farming. • Critical discussion regarding the challenges of Object Detection and Tracking. • Potential research lines to improve Object Detection and Tracking are identified. Object Detection and Tracking have gained importance in recent years because of the great advances in image and video analysis techniques and the accurate results these technologies are producing. Moreover, they have successfully been applied to multiple fields, including the agricultural domain since they offer real-time monitoring of the status of the crops and animals while counting how many are present within a field/barn. This review aims to review the current literature on Object Detection and Tracking within the field of Precision Farming. For that, over 300 research articles were explored, from which 150 articles from the last five years were systematically reviewed and analysed regarding the algorithms they implemented, the domain they belong to, the difficulties they faced, and which limitations should be tackled in the future. Lastly, it examines potential issues that this approach might have, for instance, the lack of open-source datasets with labelled data. The findings of this study indicate that Object Detection and Tracking are critical techniques to enhance Precision Farming and pave the way for robotization for the agricultural sector since they provide accurate results and insights on crop and animal management, and optimize resource allocation. Future work should focus on the optimal acquisition of the datasets prior to Object Detection and Tracking, along with the consideration of the biophysical environment of the farming scenarios. [ABSTRACT FROM AUTHOR]

• We introduce a unique dataset of UAV thermal imagery, meticulously collected under diverse heights, angles, and climatic conditions, designed to challenge and validate thermal imaging analysis. • An approach for tracking mammalian herbivores in thermal imagery using the Multi-Object Tracking and Segmentation model Track R-CNN is proposed. • 3D convolutions outperform LSTM convolutions as a temporal component. • The association head outperforms optical flow as a tracking mechanism. • The proposed approach is robust across a wide range of lighting and thermal conditions and on imagery collected at various heights and angles. Lightweight Unmanned Aerial Vehicles (UAVs) are emerging as a remote sensing survey tool for animal monitoring in several fields, such as precision livestock farming. Together with state-of-the-art computer vision techniques, UAV technology has drastically escalated our ability to acquire and analyse visual data in the field, lowering both costs and complications associated with collection and analysis. This paper addresses monitoring mammalian herbivores using the unexploited field of thermal Multi-Object Tracking and Segmentation (MOTS) in UAV imagery. In our research, a state-of-the-art MOTS algorithm (Track R-CNN) was trained and evaluated in the segmentation, detection and tracking of dairy cattle. Data collection was carried out in two farms with a UAV carrying a thermal camera at various angles and heights, and under different light (overcast/sunny) and thermal (16.5 °C range) conditions. Our findings suggest that dataset diversity and balance, especially regarding the range of conditions under which the data was collected, can significantly enhance tracking efficiency in specific scenarios. For training the algorithm, transfer learning was used as a knowledge migration method. The performance of our best model (68.5 sMOTSA, 79.6 MOTSA, 41 IDS, 100 % counting accuracy, and 87.2 MOTSP), which utilizes 3D convolutions and an association head, demonstrates the applicability and optimal performance of Track R-CNN in detecting, tracking, and counting herbivores in UAV thermal imagery under heterogenous conditions. Our findings demonstrate that 3D convolutions outperform Long-short Term Memory (LSTM) convolutions. However, LSTM convolutions also show optimal performance, offering a viable alternative. Furthermore, our results highlight the inability of Optical Flow to track motionless animals (-15 sMOTSA, −4.1 MOTSA and 2076 IDS) and the proficiency of the association head in differentiating static animals from the background. This research contributes to the growing body of knowledge in automated mammalian herbivore monitoring, with potential applications such as precision livestock farming and wildlife conservation. [ABSTRACT FROM AUTHOR]

• An anti-lodging machine, comprising an anti-lodging device and an AGV (Automated Guided Vehicle) was designed to replace manual methods. • A machine vision system was developed to aim the anti-lodging device at the crown buds. • The YOLOv5s algorithm is enhanced for detecting immature pineapple crown buds. • The mAP_0.5 of the improved YOLOv5s algorithm is 3.71 %, 1.69 %, and 1.19 % higher than the original YOLOv5s, SSD, and Faster R-CNN, respectively. • The operating speed of this intelligent anti-lodging method is 2.7 times faster than manual binding. Sunscorch, a common disease in pineapple cultivation, is caused by direct sunlight on the surface of lodging pineapple fruits. To mitigate this issue, pineapple fruits are often bagged or supported with fixing rods. This paper presents the design of an anti-lodging machine, comprising an anti-lodging device and an AGV (Automated Guided Vehicle), and develops a machine vision system to precisely aim the anti-lodging device at the crown buds. The YOLOv5s algorithm is enhanced for detecting crown buds. The improvement of the YOLOv5s algorithm includes the introduction of the attention mechanism CBAM to enhance feature extraction, the addition of BiFPN and GhostNet to strengthen feature fusion and to reduce detection time and computational cost. By transforming pixel coordinates, the relative coordinates between the crown bud and the device are calculated to achieve accurate targeting. The experiments demonstrate that the precision, recall, and map_0.5 of the improved YOLOv5s algorithm are 91.89 %, 92.25 %, 94.01 %, and the detection speed is 15.71 ms/image. The average detection speed is 1.14 ms/image and 9.45 ms/image faster than that of SSD and Faster R-CNN, benefiting from a parameter number only 3/4 and 3/7 of SSD and Faster R-CNN, respectively. The results of contrast experiments indicate that, although the lodging prevention rate of this intelligent anti-lodging method is 4 % lower than the manual binding, its operating speed is 2.7 times faster. [ABSTRACT FROM AUTHOR]

• Cows and their abdomen areas were located using YOLOv5. • Individual cows were measured remotely, non-invasively, and simultaneously. • Heat stress was detected when the respiration rate reached 60 breaths per minute. In cattle, respiration rate (RR) provides researchers and practitioners with valuable physiological information. However, traditional visual observation requires massive labour and is impractical for large-scale commercial farms. Recently developed vision-based RR measurement methods are highly limited to measuring a small number of animals in a controlled environment. Therefore, this paper aimed to propose a vision-based multi-object RR measurement method for dairy cows lying in free stalls. An RGB camera was aimed at the lying zone and was able to cover about 16 stalls. The proposed framework first utilised two YOLOv5-based networks to segment cow instances and detect cow flank objects. Next, an object tracker was used to link the predictions of each cow throughout the video clip. The Lucas-Kanade optical flow was then calculated specifically on the overlapped area of the cow mask and the flank bounding box. Finally, RR was extracted using Fast Fourier Transform. The results show that the proposed method had a precise RR measurement with a correlation coefficient of 0.944, a root mean square error of 5.35 breaths per minute, and an intraclass correlation coefficient of 0.974 when compared to visual observation. The piecewise regression models identified a change in RR when the ambient temperature reached 23.6 °C or the temperature-humidity index reached 72. The corresponding RR thresholds were 60.9 and 60.2 breaths per minute, respectively. Collectively, these results can be used to inform an automated local cooling system, e.g., fans in the lying area. However, more experiments and calibration with data collected using more cost-effective video recording systems are required before this technology can be applied on farms. [ABSTRACT FROM AUTHOR]

• Amaranthus palmeri growth stage recognition was possible with YOLO architectures. • YOLO v8-X achieved the highest mAP@[0.5:0.95] for growth stage recognition of 47.34% • A. palmeri single-class outperformed 8-class recognition by over 20% (v7-original) • The first high-quality, benchmarking, growth stage dataset is provided. Many advanced image-based precision agricultural technologies for plant breeding, field crop research, and site-specific crop management hinge on the reliable detection and phenotyping of plants across highly variable morphological growth stages. Convolutional neural networks (CNNs) have shown promise for image-based plant phenotyping and weed recognition, but the ability to recognize growth stages, often with stark differences in appearance, is uncertain. Palmer amaranth (Amaranthus palmeri) is a particularly challenging weed plant in cotton (Gossypium hirsutum) production; due to high genetic diversity, it exhibits highly variable plant morphology across growth stages over a growing season, as well as between plants at a given growth stage. This paper investigates eight-class growth stage recognition of A. palmeri in cotton as a challenging detection case study for You Only Look Once (YOLO) architectures. In total, 26 architecture variants from YOLO v3, v5, v6, v6 3.0, v7, and v8 are compared on an eight-class growth stage dataset of A. palmeri. The highest mean average precision (mAP@[0.5:0.95]) for recognition of all growth stage classes was 47.34% achieved by v8-X, with inter-class confusion across visually similar classes. With all growth stages grouped as a single class, performance increased, with a maximum mAP@[0.5:0.95] of 67.05% achieved by v7-Original. Single-class recall of up to 81.42% was achieved by v5-X, and precision of up to 89.72% was achieved by v8-X. Class activation maps (CAM) were used to understand model attention on the complex dataset. Fewer classes, grouped by visual or size features improved performance over the ground-truth eight-class dataset. Successful growth stage detection highlights the substantial opportunity for improving plant phenotyping and weed recognition technologies with open-source object detection architectures. Additionally, the first open-access, benchmarking growth stage dataset ' Palmer amaranth growth stages – 8 (PAGS8) ' is presented, which is available here: https://weed-ai.sydney.edu.au/datasets/5c78d067-8750-4803-9cbe-57df8fae55e4. [ABSTRACT FROM AUTHOR]

• A transformer-based multiple fish tracking network: TFMFT. • Multiple Association (MA) method for maximum matching. • Empirical studies on diverse transformer tracking architectures. • A multiple fish tracking dataset with various fish species and scenarios. Recent advancements in fish tracking methodologies provide valuable solutions for assessing fish growth, marine fisheries, and biological research. In particular, there has been a burgeoning interest in vision-based methods for fish tracking, owing to the enhanced computational capabilities facilitated by deep learning models. However, these methods face several challenges, including poor fish detection performance under complex backgrounds, the potential for identification switches caused by the non-rigid features and occlusions of fish, and the limited fault tolerance of extant approaches. In this paper, a t rans f ormer-based m ultiple f ish t racking model (TFMFT) is proposed, specifically designed to address the issue of instance loss of fish targets in aquaculture ponds with complex background disturbance. In particular, we introduce a Multiple Association (MA) method that bolsters fault tolerance in tracking by synthesizing simple Intersection-over-Union matching in the identification (ID) matching module. Through empirical studies across diverse Transformer-based models, we comprehensively assessed the influence of architecture design on data requirements. Furthermore, to evaluate the performance and generalizability of fish tracking models, we present the Multiple_Fish_Tracking_2022 (MFT22) dataset. The results demonstrate that TFMFT achieves 30.6% IDF1 (Identification F-Score) on the MFT22 dataset, outperforming the state-of-the-art by 10.9% and showcasing superior performance over other models. The resources and pre-trained model will be available at: https://github.com/vranlee/TFMFT. [ABSTRACT FROM AUTHOR]

• A dataset on small target pests on greenhouse sticky traps. • A novel solution was developed for counting multiple small targets based on density maps, simultaneously. • An adaptive sample equalization algorithm was developed to tackle the issue of class imbalance. Whiteflies and fruit flies are common pests that adversely affect greenhouse crops, so it is vital to control their numbers promptly. Current research involves setting up artificial monitoring devices with fixed yellow sticky traps (YSTs), for pest surveillance. However, these devices can only be installed in specific environments and additional interventions may impact the actual distribution of greenhouse pests. On the other hand, detection methods, which are the most commonly used, struggle to accurately detect overlapping or closely interacting small insects, which is a common occurrence in the presence of pests. In this paper, we propose a novel solution for the multi-species pest recognition and counting method based on a density map to tackle pest counting in the greenhouse. To reduce interference between different types of objects, we used two annotation maps to represent the distribution of each object separately. Then, two counting branches are designed to count the pests, respectively. Finally, the detected pests are mapped onto the original image to display their spatial positions. In addition, an adaptive sample equalization method is proposed to address the issue of class imbalance in images, and the atrous spatial pyramid pooling (ASPP) structure and the CBAM attention mechanism are added to the fruit fly branches to focus on learning fruit fly features from crowded edges of YSTs. When applied to split images, our method achieved reliable counting performance in counting both whiteflies and fruit flies, with R2 scores of 0.973 and 0.972, respectively. This method has been embedded into mobile devices to assist greenhouse workers in monitoring pest populations in the field. The experiments show that the proposed method provides a promising direction for counting multiple small target insects and monitoring pest populations in crowded images. At the same time, it demonstrates the potential of density map methods in various small target detection scenarios. [ABSTRACT FROM AUTHOR]

• A two-stage dense captioning model for vegetable leaf diseases is proposed. • The model can identify diseases and explain the results. • Mixed disease recognition problem solved. • The recognition ability of small lesion in the model is improved. The latest advances in deep learning technology make it possible to recognize vegetable diseases from leaf images. The existing disease recognition methods based on computer vision have shown exciting achievements in terms of accuracy, stability, and portability. However, these methods cannot provide a decision-making basis for the final results, and lack a text basis to support the users' judgement. Disease diagnosis is a risky decision. If the detection method lacks transparency, the users will not be able to fully trust the recognition results, which greatly limits the application of various recognition methods based on deep learning. Aiming at the problem of low "man–machine" credibility due to the fact that deep learning-based methods are unable to provide decision-making basis, this paper proposed a two-stage image dense captioning model named "DFYOLOv5m-M2Transformer", which can generate description sentences of visualized disease features on the basis of the recognized diseased area. Firstly, we established a target detection dataset and a dense captioning dataset containing leaf images of 10 diseases, involving 2 vegetables, i.e., cucumber and tomato. Secondly, we chose the DFYOLOv5m network as the disease detector to extract the diseased area from the image, and the M2-Transformer network as the decision basis generator to generate description sentences of disease features. Then, the Bi-Level Routing Attention module was introduced to extract fine-grained features under complex backgrounds in order to resolve the problem of poor feature extraction in case of mixed diseases. Finally, we used Atrous Convolution to expand the receptive field of the model, and fused NWD and CIoU to improve the model's performance in detecting small targets. The experimental results show that the IoU and Meteor joint evaluation indicator of DFYOLOv5m-M2Transformer achieved a mean Average Precision (mAP) of 94.7 % on the dense captioning dataset, which was 7.2 % higher than that of the best-performing model Veg-DenseCap in the control group. Moreover, the decision basis that is automatically generated by the proposed model is characterized by the advantages of high accuracy, correct grammar and large sentence variety. The outcome of this study provides a new idea for optimizing the user experience in using vegetable disease recognition models. [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]

• Thousands of weed images are acquired and medium scale dataset is built at seedling stage of cotton field scenario. • The proposed deep learning model for weed detection in cotton field at seedling stage is fast and accurate. • A vision system attached to the scene perception mechanism of intelligent robot in farmland maintains good performance. • The online weed detection and spraying experiments of the intelligent spraying robot in field condition are prospective. The precision spraying robot dispensing herbicides only on unwanted plants based on machine vision detection is the most appropriate approach to ensure the sustainable agro-ecosystem and the minimum impact of nuisance weeds. However, the coexistence of crops and weeds, the similarities of plants and the multi-scale attribute of weeds make reliable detection difficult, leading to serious limitations in the application of deep learning method to target spraying in the field environment. In this paper, 4694 representative images are acquired from cotton field scenario as the data basis for deep learning model. A novel weed detection model is constructed by employing CBAM module, BiFPN structure and Bilinear interpolation algorithm. The proposed network can effectively learn the deep information and distinguish weeds from cotton seedlings in various complicated growth states. Evaluation experiments on our constructed dataset indicate that the proposed method reaches an mAP of 98.43% with faster inference speed than Faster R-CNN. Our proposed weed detection model is also deployed in spraying robot that we developed ourselves, and field trials are conducted for detection and spraying, which could maintain the excellent performance with mAP of 97.42% and effective spraying rate of 98.93%. The ability to successfully execute the weed detection and herbicide spraying management in the field lays foundation for targeted spraying in precision weed control, which has an excellent impact on cotton cultivation and growth. [ABSTRACT FROM AUTHOR]

The detection of pests plays a crucial role in intelligent early warning systems of injurious insects and diseases in precision agriculture. However, pests strong concealment and mobility pose significant challenges to their timely detection. In this paper, we propose a novel approach called Multi-scale Dense YOLO (MD-YOLO) for detecting three typical small target lepidopteran pests on sticky insect boards. In MD-YOLO, we design three key components: the image feature extraction part, the feature fusion network, and the prediction module. To enhance the utilization of feature maps and mitigate information loss, we incorporate DenseNet blocks and an adaptive attention module (AAM) into the feature extraction part. The AAM helps capture relevant image details and improves the model's ability to exploit feature representations effectively. For effective feature integration, our feature fusion network incorporates both a feature extraction path and a feature aggregation path. This enables the deep network to leverage spatial location information from the shallower network, thereby enhancing the detection accuracy. Experimental results demonstrate the effectiveness of MD-YOLO, with detection results achieving an mAP@.5 value of 86.2%, an F1 score of 79.1%, and an IoU value of 88.1%. We conduct extensive experiments to compare MD-YOLO with state-of-the-art models, and the results showcase its superiority. Furthermore, we design an Internet of Things (IoT) system that demonstrates MD-YOLO's performance in real-world field scenes, highlighting its practical applicability. • Proposing an MD-YOLO network for the detection of 3 different species of small target pests. • Proposing a random splicing method and a random sprinkling method for image data augmentation. • Accomplishing the deployment of MD-YOLO in a pest early warning software. [ABSTRACT FROM AUTHOR]

• High accuracy in sheep multi-behavior recognition. • Propose a two-stage sheep-behavior recognition method. • Our system can recognize six types of sheep behavior. • Propose a detection model which trades off accuracy and efficiency. • The method can be used for other similar animals. Recently, computer vision has been widely used in livestock behavior detection. In animal multi-behavior detection, there is a common problem of low recognition accuracy. To address this, a deep learning-based method for sheep-behavior detection is proposed in this paper. Six types of sheep behavior can be recognized by the method. Among these behaviors, standing, feeding, and lying are normal physiological activities. Attacking, biting, and climbing are disruptive behaviors that could result in ranch losses and require immediate concern from the breeder. The method consists of two stages: the detection stage and the classification stage. In the detection stage, a detection network is used to determine whether every sheep's behavior belongs to normal physiological activities or disruptive behaviors respectively. Based on a classic network, the multi-scale feature aggregation, attention mechanism and depthwise convolution module are imported to it, which makes the network trades off between detection accuracy and model size. These improvements make the network more suitable for sheep-behavior detection. In the classification stage, the VGG network is utilized to classify the behavior of each sheep specifically. Experimental results demonstrate that the two-stage method achieves desirable results in sheep-behavior recognition. In the detection stage, the mAP of the two types of behaviors exceeds 98%. In the classification stage, the classification accuracy of all behaviors exceeds 94%. At the same time, the memory of the detection model is less than 130 MB. [ABSTRACT FROM AUTHOR]

• A convolutional neural network for detecting uneaten feed is proposed. • Attention mechanism and small object detection layer are integrated into the network. • Lightweight modules are applied to improve detection speed without loss of accuracy. • The proposed algorithm can detect residual particles accurately in real-time. In order to increase feed utilization and reduce rearing costs, it is crucial to accurately detect remaining pellets in real-time and optimize feeding mechanisms for precise feeding in intensive aquaculture. Machine vision and acoustic methods, which are widely used for the detection of residual particles, fail to adequately address scenarios such as the detection of overlapping, conjoined, faraway and dense particles, as well as particles with interference from water surface ripples and fish motion. To address these issues, this paper proposed a model for uneaten feed detection based on You Only Look Once (YOLO)-v5 network, which was built as follows: (1) The last C3 module of the backbone network is introduced into the coordinated attention (CA) mechanism, which evolves into a coordinated feature attention extraction (CFAE) module to obtain global information. (2) The GSConv module, and the VoVGSCSP module which was the aggregative mode of GSConv module, were adopted in the neck network part. (3) The shallow feature map was spliced with the deep feature map, and the tiny target detection layer (DL) was added for detection. The model elucidates the true condition of the detected residual particles by enhancing detection accuracy and minimizing model complexity, enabling precise and rapid identification of the pellet. This article employs seven sets of data to conduct a comparative analysis of the outcomes achieved before and after enhancing the three major components of the network. The experimental results showed that the YOLOv5s-CAGSDL model achieved an accuracy of 94.1% for feed particles detection, effectively solving the above problems in the actual rearing environment. Meanwhile, compared with the YOLOv5s model, AP 50 and FPS of the YOLOv5s-CAGSDL model were increased by 6.30% and 10.80% respectively, bringing potential environmental and economic benefits in the actual aquaculture environment. The source codes and datasets are available at: https://github.com/AAAArmstrong/YOLOv5s-CAGSDL. [ABSTRACT FROM AUTHOR]

• A survey of machine vision for fish tracking in aquaculture. • A taxonomy of the techniques and classify the papers based on the taxonomy. • An analysis of fish detection and tracking methods. • A review of the application of fish tracking technology in fish research. • Summarized the datasets that can be used for fish tracking. In aquaculture, Behavioral monitoring of fish contributes to scientific management and reduces the threat of loss from disease and stress. Fish tracking technology plays an important role in behavior monitoring. It can pay attention to the movement of fish at any time and discover various abnormal behaviors in time. As a non-invasive method, computer vision is a powerful tool for fish tracking.. Its tracking principle is to establish the relationship between fish positions in a continuous video sequence and get the complete movement trajectory of the fish. Nevertheless, computer vision modeling used for fish tracking is riddled with many challenges, such as fish deformation, frequent occlusion, scale change, etc. Around these difficult issues, many scholars have carried out the research. In this paper, we review the progress of tracking algorithms in fish research. Then, methods for fish tracking before deep learning are introduced. Further, a detailed discussion of fish tracking methods employing deep learning such as tracking-by-detection, deep features combined with correlation filtering methods, Siamese networks, etc. Furthermore, we summarize datasets that can be used as fish tracking and give evaluation metrics in target tracking algorithms. In addition, experimental data of several mainstream tracking algorithms on a public tracking dataset are given. Finally, we discuss the outstanding findings and look forward to the fish tracking method combined with Transformer, aiming to provide a reference for accelerating the promotion of smart fishery and precision farming. [ABSTRACT FROM AUTHOR]

• A novel occlusion-resistant instance segmentation method was developed for lactating piglets. • Predicted object centers were clustered and traced back to achieve piglet instance segmentation. • Mean average precision of 84.1 was achieved on the test set under heavy occlusion. • The predicted center is the occlusion-resistant representation of the location of an occluded object. Computer vision enables the development of new approaches to monitor the behavior, health, and welfare of animals. Instance segmentation is a high-precision method in computer vision for detecting individual animals of interest. This method can be used for in-depth analysis of animals, such as examining their subtle interactive behaviors, from videos and images. However, existing deep-learning-based instance segmentation methods have been mostly developed based on public datasets, which largely omit heavy occlusion problems; therefore, these methods have limitations in real-world applications involving object occlusions, such as farrowing pen systems used on pig farms in which the farrowing crates often impede the sow and piglets. In this paper, we adapt a Center Clustering Network originally designed for counting to achieve instance segmentation, dubbed as CClusnet-Inseg. Specifically, CClusnet-Inseg uses each pixel to predict object centers and trace these centers to form masks based on clustering results, which consists of a network for segmentation and center offset vector map, Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm, Centers-to-Mask (C2M), and Remain-Centers-to-Mask (RC2M) algorithms. In all, 4,600 images were extracted from six videos collected from three closed and three half-open farrowing crates to train and validate our method. CClusnet-Inseg achieves a mean average precision (mAP) of 84.1 and outperforms all other methods compared in this study. We conduct comprehensive ablation studies to demonstrate the advantages and effectiveness of core modules of our method. In addition, we apply CClusnet-Inseg to multi-object tracking for animal monitoring, and the predicted object center that is a conjunct output could serve as an occlusion-resistant representation of the location of an object. [ABSTRACT FROM AUTHOR]

[Display omitted] • Deep learning automates precise cereal plant head detection in agriculture. • Maize, rice, wheat, and sorghum case studies evaluated. • Comparison of deep learning approaches for object detection and image segmentation. • Challenges and opportunities for future research outlined. • Emphasis on how deep learning can contribute to increasing crop yields and food security. Cereal plant heads must be identified precisely and effectively in a range of agricultural applications, including yield estimation, disease detection, and breeding. Traditional methods that rely on manual feature extraction and thresholding take a lot of time and work, and they are also impacted by crop variability. Deep learning algorithms can be used to automate this procedure because they can directly extract complicated information from images and produce cutting-edge outcomes. This review provides a comprehensive overview of recent research on deep learning-based head detection in cereal plants, emphasizing object detection and image segmentation. We also discuss the major benefits and drawbacks of different deep learning architectures and training methods, as well as examples of their application in maize, rice, wheat, and sorghum. Developing robust image processing algorithms, using deep learning in other domains like unmanned aerial vehicles, and utilizing large and diverse datasets are all challenges outlined in our study as future research directions. Through the integration of advanced computer vision techniques with precision agriculture, this paper attempts to promote further research and innovation in this intriguing field. We provide a thorough analysis of current developments in deep learning-based head detection for cereal plants and emphasize how this technology can contribute significantly to precision agriculture. [ABSTRACT FROM AUTHOR]

• Computer vision technology can identify and classify fish with different feeding status. • In the study of fish feeding behavior, the method of combining active learning and convolutional neural network can achieve higher accuracy while reducing the number of labeled samples. • The Label-score proposed in this paper can comprehensively evaluate the accuracy and the number of labeled samples. • Automatic classification of different feeding status of fish in aquaculture. The image-based fish feeding status analysis has the potential to bring the aquaculture industry a more intelligent and precise feeding process. However, most of this type of research currently is predicated on a fully supervised approach that not only requires experienced experts for sample labeling, but also the process appears to be time consuming and resource intensive. This makes it very difficult to obtain labeled samples. To solve this problem, we proposed a model method that combines the VGG16 network with active learning to reduce the number of labeled samples needed in the analysis of fish feeding status. The algorithm is composed of two modules: feature extraction classifier and information extractor. In addition, multiple training strategies such as Mish activation function and RAdam optimizer were used in the feature extraction classifier to improve the classification performance. Optimal selection of information entropy method was used to obtain the most informative samples. To evaluate the effectiveness of our method, we designed a fish feeding experiment and verified the performance of the proposed algorithm with a special focus on the accuracy and the number of labeled samples required. The results showed that the proposed algorithm has achieved satisfying results in accuracy (the test accuracy is 98%) and the number of labeled samples (approximately 1/10 of the original dataset). Therefore, the method proposed in this paper can be applied to aquaculture production to provide a reference for farmers' feeding operations. [ABSTRACT FROM AUTHOR]

• An image restoration method based on convolutional autoencoder is proposed. • Use the area growth method to divide and code the obstruction according to the contour. • Predict the overall outline of citrus based on the unobstructed citrus outline. • Fill the area to be repaired with the selection method of similar pixels. • The repair rate and image recognition rate have been effectively improved. At present, machine vision and deep learning theories have been widely used in fruit recognition and picking. However, in the process of identification and picking, there are often situations where the target is blocked, existing methods cannot accurately identify, or the identification accuracy rate is low. For improving the recognition rate of fruits under occlusion, this paper proposes an image restoration method based on a convolutional auto-encoder. This method first encodes the obstruction and then determines the general shape of the occluded fruit and compares the general shape with the coding part is combined to determine the area to be repaired. Finally, the area to be repaired is filled with pixels to realize the image repair and recognition of the occlusion map. The average repair rate of the method proposed in this paper is 95.96%, the restoration rate is 3.52% higher than the traditional convolution method, the L2 loss value is 0.63% lower than the traditional convolution method, the average detection accuracy of the restored fruits is 94.77%. [ABSTRACT FROM AUTHOR]

• This paper reviews the research status of typical machine-vision-based weeding robots. • This paper has evaluated the performance of eight baseline methods that are based on deep learning on a public dataset. • This paper presents the application of object detection algorithms based on deep learning to weed identification. Due to its obvious advantages in saving labor and pesticides, weeding robots are one of the key technologies for modern and sustainable agriculture and have attracted increasing attention from researchers and developers. Some papers on machine-vision-based weeding robots have been published in recent years, yet there is no clear attempt to systematically study these papers to discuss the components of a robotic weed control system, such as visual navigation, weed detection and directional weeding. In this paper, typical machine-vision-based weeding robots proposed or constructed in the last 30 years, together with a few open datasets for weed detection, are reviewed. Key technologies such as image preprocessing, image segmentation, navigation line extraction, and weed recognition based on machine learning (ML) or deep learning (DL) for weeding robots are discussed. To illustrate the application of DL algorithms to weed detection, this paper provides weed object detection results and a comparative analysis of eight baseline methods based on DL using a public dataset. The study found that there are still many issues that need to be addressed in each part of the robotic weeding control system. Because of environmental variation and system complexity, machine-vision-based weeding robots are still in their early stages. The results of the systematic review provide an understanding of innovative trends in the use of machine vision in weeding systems and references for future research on weeding robots. [ABSTRACT FROM AUTHOR]

• Used an image segmentation method (GrabCut) without capturing background images to extract target objects. • Solved the over-segmentation problem by marking the segmentation area before watershed segmentation. • CNN-SVM model was better than single CNN and SVM models for recognition of multiple postures about grouped pigs. • Realized the automatic recognition of multiple postures about grouped pigs under commercial conditions using depth image. Animal posture is a manifestation of animal behavior, and an animal's behavior provides information about their health, welfare, and living environment. In recent years, machine vision and machine learning technologies have been widely used to detect individual or group behavior of pigs. The purpose of this study is to use machine vision and deep learning technologies to recognize and score multiple postures (standing, sitting, sternal recumbency, ventral recumbency and lateral recumbency) of pigs under commercial conditions based on depth images. In this study, the Azure Kinect DK depth camera with a top view was used to obtain the depth image of pigs, and the target pig image was obtained by GrabCut image segmentation and watershed segmentation of target object calibration. Then, based on the characteristics of the image, the convex hull, boundary, and the depth distance of the shoulder and the hip were obtained. The ratio of the convex hull perimeter to the boundary and the ratio of the convex hull area to the boundary, as well as the depth distance of the shoulder and the hip, and the depth distance ratio of the shoulder to the hip were obtained as the input of the Convolutional Neural Network-Support Vector Machine (CNN-SVM) classification model, and the model was trained and tested. In various classifier detection experiments, the performance of our pig posture classifier for standing posture and lateral recumbency posture was better, with the area under the receiver operating characteristic (AUC) values being 0.9969 and 0.9967, respectively. However, the performance of sitting posture, sternal recumbency posture and ventral recumbency posture classifier was slightly worse but still had good performance: AUC values were 0.9790, 0.9355 and 0.9795, respectively. The model in this article was used to detect the average postures of pigs in one day (taking the average for eight consecutive days), and it was found that the proportion of lying postures was higher than other postures (lying postures were 72%, standing postures were 20%, and sitting postures were 8%). The proportion of standing postures in the daytime was higher than that in the evening, and lying posture was the opposite. The proportion of the three lying postures also changes over time. This study compared the difference of posture recognition accuracy between the model in this paper (CNN-SVM), SVM and CNN; using the same training data and experimental data, the accuracy of posture recognition of the three models was 94.6368%, 92.2175% and 90.5396%, respectively. Therefore, the recognition accuracy of the model in this paper was improved greatly compared with CNN and SVM. [ABSTRACT FROM AUTHOR]

• A new deep learning framework to detect dairy cow lameness was proposed. • This method was more stable in the antagonism of environmental robustness. • This method improves the detection accuracy of dairy cow lameness. Manual detection of lameness poses several problems, such as difficulty in finding sudden, severe or early lameness behavior. A dairy cow's lameness is closely related to curvature of the cow's back. Focusing on the curvature features of dairy cows' backs, this study proposes a lameness detection method that combines machine vision technology with a deep learning algorithm. Firstly, the FLYOLOv3 algorithm was used to construct a Cow's Back Position Extraction (CBPE) model to realize the extraction of the dairy cow's back position coordinates. Simultaneously, a First to Last Frame Image Difference (FLFID) algorithm was used to construct a Cow's Object Region Extraction (CORE) model to separate the dairy cow from the image background and obtain the pixel region of the dairy cow. Then, a Cow's Back Curvature Extraction (CBCE) model was used to extract the dairy cow's back curvature data from the acquired dairy cow's back position and pixel region of the dairy cow. Finally, a Noise+ Bilateral Long Short-term Memory (BiLSTM) model was used to predict the curvature data and match the curvature features of the dairy cow's lameness, so as to classify and detect dairy cow lameness. To verify the effectiveness of the algorithm, 567 videos were used to train the network model in a Long Short-term Memory (LSTM) model, a BiLSTM model, Noise+LSTM model, and the model proposed in this paper, respectively, and 243 videos were used for verification and testing. According to the fitting curvature data of the dairy cows' back obtained by the algorithm used in this paper, it was found that the average classification accuracy of the model proposed in this research was 8.04%, 2.09%, and 5.78% higher than the average classification accuracy of the LSTM, BiLSTM, Noise+LSTM models, respectively. In the parallel experiment that classified the detection of dairy cow lameness, the average classification accuracy of the model proposed in this paper was 96.61%. The above results show that the lameness of dairy cows can be correctly detected through analysis of the curvature features of dairy cows' backs. The proposed method is a novel, deep learning-based method for dairy cow lameness early detection which may have significant economic impact on the dairy industry, and the proposed method provides an innovative means for detecting dairy cow lameness. [ABSTRACT FROM AUTHOR]

• Introducing a new citrus leaf diseases dataset called CCL'20 with annotations. • Introducing various recent state-of-the-art deep learning architectures for citrus disease detection. • In-depth comparison of the architectures and fine-tuning them for fast and precise detection of citrus diseases on leaves. • Accurate diagnosis of multiple instances of diseases of variable size on citrus leaves. Citrus (Citrus reticulata) plants are affected by several diseases and require keen attention to detect and cure the diseases in time; otherwise, significant financial loss is incurred. With the advancement of computer vision and deep learning techniques, identifying various diseases is becoming simpler. However, this process requires a proper dataset of infected leaves and a suitable detector to recognise the diseases. Because the publicly available citrus leaf datasets are not annotated, they are not suited for disease detection tasks. Therefore, a new dataset (called CCL'20) comprising images of infected citrus leaves with multiple classes of diseases, including precise annotations, is developed. Primarily, machine learning models are used in plant disease detection, and only limited deep learning models are utilised in agricultural applications. This paper has identified the CNN based detectors best suited for agricultural engineering, such as CenterNet, YOLOv4, Faster-RCNN, DetectoRS, Cascade-RCNN, Foveabox and Deformabe Detr, implemented and fine-tuned them to detect citrus leaf diseases using our CCL'20 dataset. Extensive performance and computational analysis is carried out to determine how effectively these models diagnose different stages of citrus leaf diseases. This paper presents the state-of-the-art CNN detectors for citrus leaf disease detection, evaluated based on their precision, recall, and other valuable parameters such as training parameters, inference time, memory usage, speed and accuracy trade-off for each model. The results show that the Scaled YOLO v4 P7 achieves fast and early prediction of the diseases, and CenterNet2 with Res2Net 101 DCN-BiFPN predicts the early stage of citrus leaf diseases with high accuracy to other recent and efficient detecting models. [ABSTRACT FROM AUTHOR]

Plant phenotyping is the study of complex plant traits to evaluate its status depending on the life-cycle conditions. Often, these evaluations are carried out by human operators, and the accuracy could be biased by their experience and skill, especially when dealing with huge amounts of data produced by high-throughput phenotyping (HTP) platforms. With the rapid development of key enabling technologies, HTP is only made possible by the vast amounts of data made available by computer vision systems. In this scenario, artificial intelligence algorithms play a key role in the automation, standardization, and quantitative analysis of large data. This paper focuses on detecting tomato plants phenotyping traits using single-stage detectors (either stand-alone or ensemble) based on YOLOv5, aiming to effectively identify nodes, fruit, and flowers on a challenging dataset acquired during a stress experiment conducted on multiple tomato genotypes. Results demonstrate that the models achieve relatively high scores, considering the particular challenges of the input images in terms of object size, similarity between objects, and their color. • YOLOv5-based deep learning models for tomato plant phenotyping. • Performance evaluation using stand alone detectors and ensemble strategies. • Identification of nodes, fruits and flowers on a challenging image dataset (different object size, similarity between objects and their color). [ABSTRACT FROM AUTHOR]

• The software can extract 54 phenotypic features of various-sized plant seeds. • These features provide efficient separation of seeds with different qualities. • Using features for modeling can realize seed clarity, vigor, and purity testing. • The established models can be applied in AIseed for visual seed quality testing. Seed quality is the key to seedling development, which affects field yield. Researchers, breeders, and processors may need to investigate seed quality on a large scale, which requires high-throughput methods, which are currently lacking. To address this bottleneck, we have developed a novel software named AIseed that automatically captures and analyzes great traits, such as the shape, color, or texture features of individual seeds, which are rarely effortlessly and stably determined by conventional measurements. The AIseed software is based on machine vision technology and enables high-throughput handling and phenotyping of various-sized plant seeds. After phenotyping for a total of 54 features, AIseed numbers each seed to facilitate further seed quality assessment or grain detection. In addition, AIseed includes modules for seed quality detection and prediction, such as seed clarity, purity, vigor, and viability testing, which are based on machine learning or deep learning models developed by analyzing the associations between seed quality indicators and the features obtained by the software. Through a series of experiments, this paper has confirmed that AIseed has a high performance in extracting phenotypic features and testing seed quality from images for seeds of several plants of different sizes, with high speed and high precision in separating seeds from the background and then analyzing them. In summary, it is a non-destructive and highly effective software for accurate seed phenotyping and seed quality assessment, which is essential for both breeding and yield improvement. [ABSTRACT FROM AUTHOR]

Image and video tools for analyzing crop scenes and plants are essential for applying precision agriculture to crop maintenance, harvesting, or pruning. In this paper, we are interested in vine pruning, a task that requires a precise understanding of the vine structure with branch type identification, orientations, and node locations. However, estimating such a structure is highly challenging, given the large variety in grapevine appearances, lighting conditions, viewpoint, the interweaving of branches, occlusions, and the level of details needed. To address these challenges, we propose ViNet: a deep-learning approach for estimating the structure of grapevine, which comprises two main steps: The first one detects nodes and identifies the branch types of the plant, as well as the spatial relation between them, whilst the second one uses the extracted nodes and branches to build a graph, out of which the structure of the grapevine is inferred. In doing so, we make four main contributions: (i) we put forward for the first time a method for automatic segmentation and extraction of the grapevine structure from images; (ii) we propose a novel approach leveraging the powerful stacked hourglass network to infer node location, branch types and the spatial relationships between them; (iii) we propose a novel shortest path weighted graph optimization step to extract connections between nodes and infer the structure, allowing to address the problem of having an unknown number of branches in the tree; (iv) we publicly release a dataset of more than 1500 grapevine images fully annotated with the structure information. Extensive experiments on this dataset demonstrate the efficiency of our approach at predicting the structure of a grapevine, achieving a precision and recall for node prediction of 95% and 90%, respectively, as well as ablation studies validating our design choices. • Deep learning-based computer vision for viticulture • Grapevine nodes, branches, and spatial relationships are estimated from a CNN • Plant structure is recovered using a graph and Dijkstra algorithm • A grapevine single Guyot dataset is publicly released [ABSTRACT FROM AUTHOR]

Volunteer cotton (VC) plants growing in the fields of inter-seasonal and rotated crops, like corn, can serve as hosts to boll weevil pests once they reach pin-head square stage (5–6 leaf stage). The VC plants therefore need to be detected, located, and destroyed or sprayed. In this paper, we present a study on using deep learning (DL) to detect VC plants in a corn field using RGB images collected with an unmanned aerial vehicle (UAV). The objectives were (i) to determine whether the YOLOv3 DL algorithm could be used for VC detection in a corn field based on UAV-derived RGB images, and (ii) to investigate the behavior of YOLOv3 on images at three different pixel scales (320 × 320, S1; 416 × 416, S2; and 512 × 512, S3). The metrics used to evaluate the results were average precision (AP), mean average precision (mAP) and F1-score at 95 % confidence level. It was found that YOLOv3 was able to detect VC plants in corn field at an average detection accuracy of more than 80 %, F1-score of 78.5 % and mAP of 80.38 %. With respect to images size, no significant differences existed for mAP among the three scales, but a significant difference was found for AP between S1 and S3 (p = 0.04) and between S2 and S3 (p = 0.02). A significant difference was also found for F1-score between S2 and S3 (p = 0.02). The overall goal of this study was to minimize boll weevil pest infestation by maximizing the true positive detection of VC plants in a corn field which is represented by the mAP values. The lack of significant differences of these at all three scales indicated that the trained YOLOv3 model can be used for VC detection irrespective of the three input image sizes. The capability of YOLOv3 to detect VC plants demonstrates the potential of DL algorithms for real-time detection and mitigation using computer vision and a spot-spray capable UAV. [ABSTRACT FROM AUTHOR]

In agriculture, aquaculture technologies such as precise feeding, fish identification and fishing based on underwater machine vision all rely on the analysis of underwater images. However, due to the scatting and attenuation of the illumination in the real-world underwater environment, turbid underwater images are inevitably degraded, limiting their applicability in many vision tasks. In this paper, we present an unsupervised deep learning framework, called Underwater Loop Enhancement Network (ULENet), to improve the quality of turbid underwater images. We first propose an underwater dataset construction scheme and construct the dataset on which the network proposed above is trained. The underwater dataset contains images of three different scenes: lake and reservoir scene data (no label), pool scene data (weakly correlated label), and laboratory scene data (strongly correlated label). Then we propose a loop enhancement structure that uses the approximate candidates as labels and improves the visual quality of the image through the iterative training process. We formulate a new underwater visual perception loss function that evaluates the perceptual image quality based on its color, contrast, saturation and clarity. During the training process, a more realistic, higher-contrast, and clearer underwater image is gradually generated. Qualitative and quantitative evaluations show that the proposed method can effectively enhance image clarity. Moreover, the enhanced images are applied to several vision tasks to achieve better results, such as edge detection, key point matching, fish target detection and saliency prediction etc. • Underwater Loop Enhancement Network (ULENet), to improve the quality of turbid underwater images. • Lake and reservoir scene data, pool scene data, and laboratory scene data. • Our method can effectively improve the visual perception on the above datasets. [ABSTRACT FROM AUTHOR]

• Computer vision methods for livestock behaviour recognition were evaluated. • Journey from computer vision to deep learning was elaborated. • This paper involves with segmentation, identification and behaviour recognition. • Research trend of livestock behaviour recognition was illustrated from four aspects. • This review could provide researchers and producers with technical references. The increasing demand for sustainable livestock products also demands new considerations in animal breeding. Breeding programs are now seeking to integrate animal behavioural phenotypes, as these relate to the productivity, health and welfare of the animals and thereby can influence yield and economic benefits in the industry. Traditional manual observation of pig behaviour is time-consuming, laborious, subjective, and difficult to achieve in continuous and large-scale operations. It is not surprising that computer vision technology with the advantages of being objective, non-invasive and continuous has been widely researched for its use in the recognition of livestock behaviours over recent years. Nevertheless, in studies of livestock behaviour recognition, computer vision technology faces some challenges, e.g., complex scenes, variable illumination, occlusion, touching and overlapping between livestock, which has limited the fast translation of technology to industry. On the other hand, deep learning technology has proven to solve these difficulties to a certain extent and is being adopted to recognise livestock behaviours. This paper mainly evaluates the recent developments in computer vision methods for recognition of these behaviours in pigs and cattle. The focus on these species is made possible by the number of studies exist quantifying behaviours that are of importance for their health, welfare and productivity such as aggression, drinking, feeding, lameness, mounting, posture, tail-biting and nursing. This review paper especially analyses the development of image segmentation, identification and behaviour recognition using tradition computer vision and more recent deep learning methods, and evaluates the evolution of key research in the field. We elaborate the research trend of livestock behaviour recognition from four aspects, i.e., development of robust livestock identification algorithms, recognition of livestock behaviours for different growth stages, further quantification of the results of behaviour recognition, and building evaluation system of growth status, health and welfare. [ABSTRACT FROM AUTHOR]

• Method of combining alternate circumrotating mechanisms and deep learning. • Quantitative summary of defects in soybean seeds at different scales. • Model comparison, selection and optimization. • Full-surface recognition rate of defects in different scales exceeds 97.7% • Realize real-time classification of soybean seeds in the edge device. Accurately sorting high-quality soybean seeds is a key task in increasing soybean yield in the breeding industry. At present, sorting systems based on machine vision focus on the recognition of only one side surface. This paper designs and develops a sorting system based on deep learning that can recognize the full surface of soybean seeds. An alternate circumrotating mechanism is used to expose the full surface feature information of the seeds, and a deep learning model is applied for accurate seed classification. We divide soybean seeds into six categories. Images are collected and masked in three brightness environments and six surfaces, and a defect scale of soybean seeds is quantified. We compare and test seven CNN models and improve the model with the best overall performance. Visualization technology is used to assess the recognition performance of different models for soybean seed defects, and the model is optimized based on the results to achieve accurate classification of seed defects at different scales. The testing process indicates that all the models have the highest accuracy under medium brightness conditions. The classification accuracy of the MobileNetV2-improved model reaches 97.84% in the masked dataset and has an inference speed of 35 FPS with NVIDIA's Jetson Nano board, realizing real-time recognition of the soybean full surface. The sorting system proposed in this paper can achieve high-precision and low-cost application, with a total sorting accuracy of 98.87% and a sorting speed of 222 seeds per minute. This method can be used as an effective tool for precise sorting of soybean seeds. Moreover, it provides an approach for full-surface detection of defective ellipsoid seeds of different scales. [ABSTRACT FROM AUTHOR]

• Proposed interdisciplinary terms and concepts in precision management of animals. • Sources of uncertainty in animal behavior labeling for computer vision applications. • Directions of advanced computer vision algorithms in pig production facilities. • Image dataset construction workflows for computer vision algorithm development. Surveillance and analysis of behavior can be used to detect and characterize health disruption and welfare status in animals. The accurate identification of changes in behavior is a time-consuming task for caretakers in large, commercial pig production systems and requires strong observational skills and a working knowledge of animal husbandry and livestock systems operations. In recent years, many studies have explored the use of various technologies and sensors to assist animal caretakers in monitoring animal activity and behavior. Of these technologies, computer vision offers the most consistent promise as an effective aid in animal care, and yet, a systematic review of the state of application of this technology indicates that there are many significant barriers to its widespread adoption and successful utilization in commercial production system settings. One of the most important of these barriers is the recognition of the sources of errors from objective behavior labeling that are not measurable by current algorithm performance evaluations. Additionally, there is a significant disconnect between the remarkable advances in computer vision research interests and the integration of advances and practical needs being instituted by scientific experts working in commercial animal production partnerships. This lack of synergy between experts in the computer vision and animal health and production sectors means that existing and emerging datasets tend to have a very particular focus that cannot be easily pivoted or extended for use in other contexts, resulting in a generality versus particularity conundrum. This goal of this paper is to help catalogue and consider the major obstacles and impediments to the effective use of computer vision associated technologies in the swine industry by offering a systematic analysis of computer vision applications specific to commercial pig management by reviewing and summarizing the following: (i) the purpose and associated challenges of computer vision applications in pig behavior analysis; (ii) the use of computer vision algorithms and datasets for pig husbandry and management tasks; (iii) the process of dataset construction for computer vision algorithm development. In this appraisal, we outline common difficulties and challenges associated with each of these themes and suggest possible solutions. Finally, we highlight the opportunities for future research in computer vision applications that can build upon existing knowledge of pig management by extending our capability to interpret pig behaviors and thereby overcome the current barriers to applying computer vision technologies to pig production systems. In conclusion, we believe productive collaboration between animal-based scientists and computer-based scientists may accelerate animal behavior studies and lead the computer vision technologies to commercial applications in pig production facilities. [ABSTRACT FROM AUTHOR]

• An overview of generative adversarial network (GAN) architectures is presented. • The application of GANs in agriculture is reviewed. • The challenges and opportunities of GANs are discussed. In agricultural image analysis, optimal model performance is keenly pursued for better fulfilling visual recognition tasks (e.g., image classification, segmentation, object detection and localization), in the presence of challenges with biological variability and unstructured environments. Large-scale, balanced and ground-truthed image datasets are tremendously beneficial but most often difficult to obtain to fuel the development of highly performant models. As artificial intelligence through deep learning is impacting analysis and modeling of agricultural images, image augmentation plays a crucial role in boosting model performance while reducing manual efforts for image collection and labelling, by algorithmically creating and expanding datasets. Beyond traditional data augmentation techniques, generative adversarial network (GAN) invented in 2014 in the computer vision community, provides a suite of novel approaches that can learn good data representations and generate highly realistic samples. Since 2017, there has been a growth of research into GANs for image augmentation or synthesis in agriculture for improved model performance. This paper presents an overview of the evolution of GAN architectures followed by a first systematic review of various applications in agriculture and food systems (https://github.com/Derekabc/GANs-Agriculture), involving a diversity of visual recognition tasks for plant health conditions, weeds, fruits (preharvest), aquaculture, animal farming, plant phenotyping as well as postharvest detection of fruit defects. Challenges and opportunities of GANs are discussed for future research. [ABSTRACT FROM AUTHOR]

• A comprehensive benchmark of deep learning models was built for weed identification. • A dataset was created with 5187 images for 15 weed classes collected in cotton fields. • Weighted loss function was effective for improving accuracies for minority weed classes. • Deep learning-based cosine similarity was useful for the interpretation of classifications. • Both software programs and the dataset for weed identification are publicly accessible. Precision weed management offers a promising solution for sustainable cropping systems through the use of chemical-reduced/non-chemical robotic weeding techniques, which apply suitable control tactics to individual weeds or small clusters. Therefore, accurate identification of weed species plays a crucial role in such systems to enable precise, individualized weed treatment. Despite recent progress, the development of a robust weed identification and localization system in the presence of unstructured field environments remains a serious challenge, requiring supervised modeling using large volumes of annotated data. This paper makes a first comprehensive evaluation of deep transfer learning (DTL) for identifying common weed species specific to cotton (Gossypium hirsutum L.) production systems in southern United States (U.S.). A new dataset for weed identification was created, consisting of 5187 color images of 15 weed classes collected under natural light conditions and at varied weed growth stages, in cotton fields (primarily in Mississippi and North Carolina) during the 2020 and 2021 growth seasons. We evaluated 35 state-of-the-art deep learning models through transfer learning with repeated holdout validations and established an extensive benchmark for the considered weed identification task. DTL achieved high classification accuracy of F1 scores exceeding 95%, requiring reasonably short training time (less than 2.5 h) across models. ResNeXt101 achieved the best overall F1-score of 98.93 ± 0.34%, whereas 10 out of the 35 models achieved F1 scores near or above 98.0%. However, the performance on minority weed classes with few training samples was less satisfactory for models trained with a conventional, unweighted cross entropy loss function. To address this issue, a weighted cross entropy loss function was adopted, which achieved substantially improved accuracies for minority weed classes (e.g., the F1-scores for Xception and MnasNet on the Spurred Anoda weed increased from 48% to 90% and 50% to 82%, respectively). Furthermore, a deep learning-based cosine similarity metric was employed to analyze the similarity among weed classes, assisting in the interpretation of classifications. Both the codes (https://github.com/Derekabc/CottonWeeds) for model benchmarking and the weed dataset (https://www.kaggle.com/yuzhenlu/cottonweedid15) of this study are made publicly available, which expect to be a valuable resource for future research on weed identification and beyond. [ABSTRACT FROM AUTHOR]

[Display omitted] • Plant stand count methods using unmanned aerial vehicle imagery were reviewed. • Computer vision methods produced promising results in plant stand count analysis. • Several research gaps were identified and recommendations provided. • Results will be helpful to stakeholders like farmers, producers, and researchers. Plant stand count helps in estimating the yield and evaluating the planter's efficiency and seed quality. Traditional methods of counting the plants by manual measurement are time-consuming, laborious, and error-prone. In contrast, the ground-based sensing methods are limited to smaller spaces. High spatial resolution images obtained from unmanned aerial vehicles (UAV) can be used in conjunction with computer vision algorithms to evaluate plant stand count, as it directly influences the yield. In spite of the importance of high-throughput plant stand count in row crop agriculture, no synthesized information in this specific subject matter is available. Therefore, the objective of this paper was to perform a systematic literature review of the current studies that focus on evaluating plant stand count using UAV imagery to provide well-synthesized information, identify research gaps, and provide suitable recommendations. In this study, a comprehensive literature search was performed on three academic databases (Agricola, Web of Science, and Scopus), and a total of 29 articles were found based on search terms and selection criteria for review. From the systematic review, it can be concluded that: an appropriate stage after plant emergence without canopy overlap is necessary for image acquisition; optimal flying height should be selected to balance the field coverage and accuracy; L*a*b* color space can provide better segmentation; hyperspectral camera imagery can provide good discrimination; deep learning with data augmentation and transfer learning models can be used to reduce the computational time and resources; the stand count methodology that is successful with corn and cotton could be extended to other row crops and horticultural crops; and application of direct image processing and use of open-source platforms is required for stakeholder participation. The review will be helpful to the farmers, producers, and researchers in selecting and employing the UAV-based algorithms for evaluating plant stand count. [ABSTRACT FROM AUTHOR]

• A method for measuring the body size of live chickens. • The accuracy and stability of measurement are greatly improved. • Reduce the image interference from live chicken feathers. • Reduce the image interference from live chicken feathers. • YOLOv5 used to improve measurement speed. The phenotypic parameters of breeder chickens are the external expression of their genetic genes. Accurate and efficient acquisition of chicken phenotypic parameters is a key technology in poultry breeding. However, the traditional manual measurement method is time-consuming and laborious, and the measurement error is large, which seriously affects the effect and efficiency of poultry breeding. Therefore, in this paper, Xinghua chicken as the research object, combining computer vision technology and deep learning algorithm, proposed the shank length and circumference measurement algorithm (SLCM) of breeder chickens. The algorithm firstly detected and extracted the region of chicken claw and shank bottom base by YOLOV5 algorithm. And then extracted the key points of shank length and circumference measurement based on the pixel distribution features of row direction and column direction in this region, and finally calculated the values of shank length and circumference. The test results of SLCM algorithm show that compared with the traditional manual measurement method, the Pearson correlation coefficient of shank length and circumference measured by algorithm is 0.8722 and 0.8869. Besides, this algorithm not only has smaller standard deviation (standard deviation of shank length is 1.35 mm, and shank circumference is 0.25 mm) and coefficient of variation (the mean coefficient of variation of shank length is 0.02, maximum is 0.06; and the mean coefficient of variation of shank circumference is 0.01, maximum is 0.03), but also has higher measurement efficiency. This research provided reference and basis for the accurate measurement of phenotypic parameters of breeder chickens and had important significance for poultry breeding. [ABSTRACT FROM AUTHOR]

• This research focuses on soybean cultivar identification which is known as a very fine-grained classification problem. • The first attempt on fusing deep features of triplet leaf image patterns is reported. • Two deep feature fusion methods, distance fusion and classifier fusion, are proposed. • A novel fine-tuning strategy that addresses the issue of few-shot soybean leaf image classification is designed. • The proposed method achieves an exciting accuracy of 83.55% on classifying 200 soybean cultivars. Soybean cultivar recognition plays a vital role in cultivar evaluation, selection and production. Recently, there is an increasing interest in taking leaf image patterns as clues for distinguishing soybean cultivars. However, due to the higher inter-class similarity of soybean cultivars over plant species, the cultivar classification accuracies reported by the existing methods are far lower than those published on plant species recognition which make computer vision community have a concern whether leaf image patterns can provide sufficient discriminative information for identifying soybean cultivars. In this paper, we explore fusing deep learning features of leaves from different parts of soybean plants for achieving an accurate cultivar recognition. In our method, the deep learning features of triplet leave image patterns that consists of leaves from the lower, middle, and upper parts of soybean plants are fused by two methods, distance fusion and classifier fusion. In the former, the L 1 distance measurements defined on the deep feature spaces of triplet leaf image patterns are fused prior to using 1NN classifier for classification. While in the later, the SVM classifiers trained by the deep features of triple leaf image patterns are combined by sum rule for cultivar prediction. We use the SoyCultivar200 leaf dataset which consists of 6000 samples from 200 soybean cultivars as benchmark. Our method achieves an exciting classification rate of 83.55% which demonstrates that our proposed fusion of deep features of triplet leaf image patterns can provide strong discriminative information for accurately identifying soybean cultivars. [ABSTRACT FROM AUTHOR]

• Machine-vision system for animal identification. • Deep learning algorithms were used for sheep identification. • ArcFace loss function used for animal identification. • Sheep identification based on facial images. This paper describes a sheep biometric identification system based on facial images. A machine vision system and deep learning model were developed and applied for animal identification. The system included two 8-MegaPixels cameras installed in a controlled water trough adapted to work with NVIDIA Jetson Nano-embedded system-on-module (SoM). Data from 81 Assaf breed sheep, aged two to three months, from two different groups of sheep, were collected over a period of two weeks. The biometric identification model included two steps: face detection and classification. In order to locate and localize the sheep's face in an image, the Faster R-CNN deep learning object detection algorithm was applied. The detected face was provided as input to seven different classification models. Different transfer learning methods were examined. The best performance was obtained using a ResNet50V2 model with the state-of-art ArcFace loss function. The identification system resulted in average accuracies of 95% for the two groups tested. When applying transfer learning methods, average identification accuracies improved to 97% in both groups, and the training process was accomplished in half the time. The newly developed system proves the feasibility of individual biometric identification of sheep on commercial farms. [ABSTRACT FROM AUTHOR]

• A novel cattle face identification framework for practical production scenarios. • RetinaFace-mobilenet is designed for cattle face detection and location. • ArcFace Loss is combined with RetinaFace-mobilenet applied for individual identification. • The framework outperforms other competing algorithms and achieves accuracy of 91.3%. Cattle identification is crucial to be registered for breeding association, food quality tracing, disease prevention and control and fake insurance claims. Traditional non-biometrics methods for cattle identification is not really satisfactory in providing reliability due to theft, fraud, and duplication. In this study, a computer vision technique was proposed to facilitate precision animal management and improve livestock welfare. This paper presents a novel face identification framework by integrating light-weight RetinaFace-mobilenet with Additive Angular Margin Loss (ArcFace), namely CattleFaceNet. RetinaFace-mobilenet is designed for face detection and location, and ArcFace is adopted to strengthen the within-class compactness and also between-class discrepancy during training. Experiments on real-word scenarios dataset prove that RetinaFace-mobilenet achieves superior detection performance and significantly accelerates the computation time against RetinaNet. Three loss functions utilized in human face recognition combined with RetinaFace-mobilenet are compared and results indict that the proposed CattleFaceNet outperforms others with identification accuracy of 91.3% and processing time of 24 frames per second (FPS). This research work demonstrates the potential candidate of CattleFaceNet for livestock identification in real time in practical production scenarios. [ABSTRACT FROM AUTHOR]

• Mechanical weeding module for precise hoeing in vegetable farming. • Real-time precise stem detection in RGB images on embedded computer. • Deep learning object detection based neural network for precise stem location. • Tracking algorithm for temporal aggregation of stems in successive frames. • F1-score of 94.74% and 93.82% for maize and bean with sub-centimetric location accuracy. Developing alternatives to the chemical weeding process usually carried out in vegetable crop farming is necessary in order to reach a more sustainable agriculture. However, a precise mechanical weeding requires specific sensors and advanced computer vision algorithms to process crop and weed discrimination in real-time. In this paper we propose an algorithm able to detect, locate, and track the stem position of crops in images which is suitable for precision actions in vegetable fields such as mechanical hoeing within crop rows. The algorithm is twofold: (i) a deep neural network for object detection is first used to detect crop stems in individual RGB images and then (ii) an aggregation algorithm further refines the detections taking advantage of the temporal redundancy in consecutive frames. We evaluated the pipeline on images of maize and bean crops at an early stage of development, acquired in field conditions with a camera embedded in an experimental mechanical weeding system. We reported F1-scores of respectively 94.74% and 93.82% with a location accuracy around 0.7 cm when compared with human annotation. Moreover, this pipeline can operate in real-time on an embedded computer consuming as little power as 30 W. [ABSTRACT FROM AUTHOR]

• An algorithm for detecting ripe fruit under natural illumination and occlusion conditions is proposed. • A graying factor K is proposed to remove the complex background of colorful ripe fruits. • Improving the Hough Transform by selecting equidistant points and increasing judgement threshold. • A real contour Hough Transform was proposed to detect ripe fruits or occluded circles. For an accurate detection of ripe fruits under uneven illumination and ubiquitous occlusion conditions, this paper proposes a method to detect and locate ripe fruits based on machine vision. There are four key steps in this method including image graying and background removal, binary image optimization, true contour fragments extraction, and fruit fitting. For testing the proposed method, field experiments were conducted with tomato and citrus, and the ripe fruits in complex environments were successfully detected and located. From the detection experiments, it showed that the recognition rate for ripe fruits in the near zone of the proposed method was higher than 97.44%, the average time consumption was 0.2966 s, and the positioning error was less than 4.41%. In addition, it can be concluded from the comparative experiment that the proposed method is superior to conventional Hough transform, random Hough transform, and other methods based on deep learning in terms of detection rate, time performance and positioning accuracy. Therefore, it can be applied to picking robots for real-time detecting and locating ripe fruits. [ABSTRACT FROM AUTHOR]

• Locate and classify broccoli heads based to its state of maturity. • The proposed system can be adapted, being able to work directly with natural light. • The cost of the required computational equipment is quite low for agriculture applications. Broccoli is a vegetable grown worldwide due to its good nutritional properties. The harvest of this product is done selectively by hand depending on their size and state of maturation for both fresh market and agri-food industry. The final aim of our work is the development of a machine that is able to automatically harvest only those broccoli heads that have the size and ripeness suitable for the agri-food industry, besides discarding those overripe or with diseases. One critical element in such a machine is a vision system that locates and classifies the broccoli heads present in photographic images, to trigger later a cutting module. In this paper, we present an approach to that vision system, based on deep learning techniques. The proposed algorithm, running in a relatively cheap hardware, is able to work in real time, locating broccoli heads in 640 × 480 px digital images, and classifying then into harvestable , immature and wasted classes. Tested with images taken in real conditions, with many heads partially hidden by leaves, the system was able to correctly locate and classify up to 97% of the cases presented in the test set. [ABSTRACT FROM AUTHOR]

• Dual attention feature pyramid was developed to instance segmentation of pigs. • Dual attention blocks were better than other state-of-art attention modules. • Visualization of attention maps suggested that it can capture semantic similarity. • Prediction of public datasets demonstrated the practicability of attention blocks. In respect of pig instance segmentation, the application of traditional computer vision techniques is constrained by sundries barrier, overlapping, and different perspectives in the pig breeding environment. In recent years, the attention-based methods have achieved remarkable performance. In this paper, we introduce two types of attention blocks into the feature pyramid network (FPN) (see nomenclature table) framework, which encode the semantic interdependencies in the channel (named channel attention block (CAB)) (see nomenclature table) and spatial (named spatial attention block (SAB)) (see nomenclature table) dimensions, respectively. By integrating the associated features, the CAB selectively emphasizes the interdependencies among the channels. Meanwhile, the SAB selectively aggregates the features at each position through a weighted sum of the features at all positions. A dual attention block (DAB) (see nomenclature table) is proposed to integrate CAB features with SAB information flexibly. A total of 45 pigs with 8 pens are captured as the experiment subjects. In comparison with such state-of-art attention modules as convolutional block attention module (CBAM) (see nomenclature table), bottleneck attention module (BAM) (see nomenclature table), and spatial-channel squeeze & excitation (SCSE) (see nomenclature table), embedding DAB can contribute to the most significant performance improvement in different task networks with distinct backbone networks. Especially with HTC-R101-DAB (hybrid task cascade) (see nomenclature table), the best performance is produced, with the AP 0.5 (average precision) (see nomenclature table) AP 0.75 , AP 0.5:0.95 , and AP 0.5:0.95-large reaching 93.1%, 84.1%, 69.4%, and 71.8%, respectively. Also, as indicated by ablation experiments, the SAB contributes more than CAB. Meanwhile, the predictive results appear a trend of increasing initially and decreasing afterwards after different numbers of SAB are merged. Besides, as revealed by the visualization of attention maps, attention blocks can extract regions with similar semantic information. The attention-based models also produce outstanding segmentation performance on public dataset, which evidences the practicability of our attention blocks. Our baseline models are available 1 1 https://github.com/zhiweihu1103/pig-instance-segmentation. [ABSTRACT FROM AUTHOR]