Search

• We proposed a delineating field parcel model (TSANet) based on satellite image time series. • TSANet learn the relevance of spatial-spectral-temporal feature representation. • TSANet is robust across space and time. • TSANet performs better than pervious methods. • The time series data from the main growing period has a greater impact. Satellite image time series (SITS), such as Sentinel-2 imagery, plays a crucial role in the delineation of agricultural fields by reducing the impacts of ambiguities due to the spatial arrangement of field boundaries. Existing delineate field parcel models rely extensively on spatial features derived from single-date imagery. However, several studies have exploited the potential of SITS to effectively tackle the complexities associated with the intrinsic consistency between agricultural fields and their boundaries. This paper proposes a novel Two-Stream Attention convolutional neural network (TSANet) to capture the subtle difference between agricultural fields from SITS. Specifically, a field temporal semantic stream is introduced to adaptively leverage the significance of spatial-spectral-temporal feature representation associated with the location of agricultural parcels, especially where transitions in crop types take place. Considering the consistency between field parcels and their boundaries, we developed a field boundary prediction stream to enhance the extraction of edge features, particularly for the extraction of small and irregular agricultural parcels. Moreover, a field parcel refining block is employed to further enhance the geometric accuracy of agricultural fields. We conducted experiments on Sentinel-2 images from the Netherlands. Results showed that our approach produced a better layout of agricultural fields, with an average F1-score of 0.91 than the existing 3D-UNet, U-TEA, and BiConvLSTM. In addition, through the analysis of both quantitative and qualitative results, the stronger robustness of the model compared to other algorithms has been verified by temporal transfer and large-scale spatial prediction. We compared the difference between SITS and the corresponding composite images, which further verified the influence of temporal variation on the proposed approach. This paper provides a general guide for delineating agricultural parcels using SITS. [ABSTRACT FROM AUTHOR]

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

• A multi-layer perceptron model for straight-turn binary classification training using K-Fold cross validation is proposed. • A convolutional neural network-based model is proposed to proform turning classification between Omega turn and T turn. • The proposal of line and turn segmentation algorithms and turn type recognition algorithms provides new methods for estimating the efficiency of agricultural machinery operations. GPS technology is an indispensable technique in "precision agriculture". In the field of path planning, the evaluation of planned trajectories is based on GPS data. However, due to the unstructured operating environment, the existing methods for evaluating the efficiency of actual operating trajectories are not yet fully developed. In the context of agricultural machinery GPS trajectories, headland turning of the field is considered a non-working segment, which related to evaluation standard for trajectory planning efficiency in many studies. This paper proposes a processing method that evaluates GPS trajectories efficiency. A turning extraction algorithm based on changes in the angle of the agricultural machine's driving direction is proposed. The turning parts of the raw GPS data are preliminarily extracted, and a method for constructing a turning extraction dataset is presented. Then a multi-layer perceptron model for straight-turn binary classification training using K-Fold cross validation is designed and applied, which can implement robust segmentation of agricultural machinery trajectory points in actual environment. Two methods for building different turning classification datasets are proposed, and turning classification is performed using two deep learning models. A convolutional neural network-based model classifies headland turnings by their normalized sequences' visualization images, while a long short-term memory network-based model utilizes the trajectory point sequences. The CNN-based model performs relatively better than the LSTM-based model. Finally, some commonly used GPS trajectory efficiency evaluation standards in the field of agricultural machinery path planning are combined to evaluate the efficiency of actual GPS trajectories of agricultural machinery. [ABSTRACT FROM AUTHOR]

Aiming at the problem that unstable identification results and weak generalization ability in feature extraction based on manual design features in weed identification, this paper take the soybean seedlings and its associated weeds as the research object, and construct a weed identification model based on K-means feature learning combined with Convolutional neural network. Combining advantages of multilayer and fine-turning of parameters of the convolutional neural network, this paper set k-means unsupervised feature learning as pre-training process, and replaced the random initialization weights of traditional CNN parameters. This method make the parameters can be obtained more reasonable values before optimization to gain higher weed identification accuracy. The experimental results show that this method with K-means pre-training achieved 92.89% accuracy, beyond 1.82% than convolutional neural network with random initialization and 6.01% than the two layer network without fine-tuning. Our results suggest that identification accuracy might be improved by fine-tuning of parameters. [ABSTRACT FROM AUTHOR]

• An end-to-end rice varieties hyperspectral classification model based on hybrid CNN (3D-CSAM-2DCNN) was proposed. • Hyperspectral dataset of 14 rice varieties were collected by UAV platform, with clipping and manually labeling. • The channel and spatial attention modules were used to capture band correlation and extract deeper spatial features. • Patch size and the number of training samples were adjusted to guarantee the performance of model. Rice is an indispensable food crop for human beings. Rice varieties are closely related to disease resistance, insect resistance, lodging resistance, grain quality and yield. Different varieties of rice have similar appearance traits and change trends, which are difficult to distinguish. It is of great significance to classify and identify rice varieties with high precision in a wide range by objective and non-destructive detection methods. In this paper, the canopy hyperspectral images of rice varieties were obtained by using the S185 hyperspectral imaging device mounted on a UAV platform. And the spectral and spatial features of 14 rice varieties were automatically learned and deeply extracted by hybrid convolutional neural network structure. In addition, in order to improve the performance of the model, the article attempts to optimize the model with the end-to-end trainable attention module. Finally, extensive experiments are carried out to prove the validity of the model. Compared with the advanced methods, the 3D-CSAM-2DCNN proposed in this paper performed the best classification on fine classification of rice varieties. The overall accuracy of 98.93% and the accuracy of more than 98.22% for single variety has been achieved. The proposed model is conducive to automatic identification of fields and crop phenotypes research, and contributes new possibilities to promote the development of precision agriculture and smart agriculture. [ABSTRACT FROM AUTHOR]

• The Transformer can effectively suppress the interference of complex background information on disease recognition. • The overlapping patch embedding improves the local continuity of the input image information of Vision Transformer. • We build a scalable and efficient identification model ConvViT to identify kiwifruit disease in complex environments with fewer parameters and computation. • ConvViT has good generalization and is a potential backbone network for feature extraction. The complex background of disease images and the small contrast between the disease area and the background easily confuse, seriously affecting the robustness and accuracy of kiwifruit disease identification models. To address the above problems, this paper proposes a disease identification model based on Vision Transformer and Convolutional Neural Network, ConvViT(Convolutional Neural Network and Vision Transformer), to identify diseases by extracting effective features of kiwifruit disease spots. The proposed ConvViT includes convolutional structure and Transformer structure: The convolutional structure is used to extract the global features of the image, and the Transformer structure is used to obtain the local features of the disease area to help the CNN see better. Meanwhile, the paper designs different models according to the number of parameters and FLOPs (floating-point operations) to improve the model's scalability. The model variants of different sizes are designed to be lightweight to run on devices with different resource constraints. We achieved 98.78% identification accuracy on the self-built kiwifruit disease dataset, with up to 4.53% improvement in identification accuracy compared to the same level of Resnet, ViT, and ResMLP, and more than 10% reduction in the number of parameters and FLOPs. Experimental results on the PlantVillage dataset and the AI Challenger 2018 also show that ConvViT has good generalizability, indicating that the proposed model can solve kiwifruit disease identification problems in complex environments and be valuable a backbone network for other identification tasks with practical applications. [ABSTRACT FROM AUTHOR]

• Proposing an end-to-end noise-tolerant feature fusion network to integrate outdoor RGB-D information. • Designing an attention-based fusion module to eliminate the negative influence of depth noise and make full use of the complementary multi-modal features. • Learning object position from multi-scale features from color images and fusion modules, further improving the network noise immunity. • Achieving accurate fruit detection in outdoor scenes with a real-time speed. In the process of farm automation, fruit detection is the basis and guarantee for yield prediction, automatic picking, and other orchard operations. RGB images can only obtain the two-dimensional information of the scene, which is not sufficient to effectively distinguish fruits that are dense growth and occlusion by branches and leaves. With the development of depth sensors, using RGB-D images with more complementary information can boost the performance of fruit detection. However, due to the nature of sensors and scene configurations, the quality of outdoor depth images is poor, posing a challenge when fusing RGB-D features. Therefore, this paper proposes an end-to-end RGB-D object detection network, termed as noise-tolerant feature fusion network (NT-FFN), to utilize the outdoor multi-modal data properly and improve the detection accuracy. Specifically, the NT-FFN first uses two structurally identical feature extractors to extract single-modal (color and depth) features, which is the base of the subsequent feature fusion. Then, to avoid introducing too much depth noise and focus the perception on the important part of the features, an attention-based fusion module is designed to adaptively fuse the multi-modal features. Finally, multi-scale features from the color images and the fusion modules are used to predict object position, which not only improves the network's ability to detect multi-scale objects but also further enhances the noise immunity of the network. In addition, this paper constructs an RGB-D citrus fruit dataset, which contributes to comprehensively evaluating the proposed network. Evaluation metrics on the dataset show that the NT-FFN achieves an AP50 of 95.4% with a real-time speed, which outperforms single-modal methods, common multi-modal fusion strategies, and advanced multi-modal detection methods. The proposed NT-FFN also achieves excellent detection results in other fruit detection tasks, which verifies its generalization ability. This study provides the possibility and foundation for performing multi-modal information fusion in outdoor fruit detection. [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]

• Proposed a cattle face recognition model based on a two-branch CNN. • Constructed face image datasets of dairy cows and beef cows. • Reduced the influence of cow face posture on recognition accuracy. • The network model is small and the recognition accuracy is high. Due to changes in cattle posture and different shooting angles, some features of collected cattle face images are missing, which leads to a decline in the accuracy of cattle face recognition. This paper proposes a cattle face recognition model based on a two-branch convolutional neural network (TB-CNN). The collected two cattle face images from different angles are input to the convolutional neural network of different channels for feature extraction, the features of the two channels are feature fused, and the global average pooling layer is combined with the classifier to identify the individual cattle. The squeeze-and-excitation block (SE) is embedded in the feature extraction network in the network model to improve the network feature extraction capability. The global average pooling layer is used to replace the fully connected layer, which improves the network classification capability and reduces the number of network parameters. The experimental results show that the recognition rate of the cattle face recognition model based on the TB-CNN is 99.85% on the Simmental beef cattle face image dataset, 99.81% on the Holstein cow face image dataset, and 99.71% on the beef cattle and cow mixed dataset. The cattle face recognition model proposed in this paper has good robustness and generalization ability, which can effectively reduce the influence of cattle face angle changes on the cattle face recognition rate and improve the accuracy of cattle face recognition. [ABSTRACT FROM AUTHOR]

• We introduce a goat image dataset for face detection and recognition. • Presenting an automatic goat face and facial landmark extraction from source image along with face alignment. • We report goat face recognition using convolutional neural network approach with 96.4% accuracy. Automatic identification of individual animals is an important step towards achieving accurate breeding histories, significant contributions to breeding and genetic management programmers. Currently, a different type of tags, tattoos, paint brands and microchips are used to uniquely identify livestock animals. However, the manual identification system is time-consuming, expensive and unreliable. In this paper, we present a deep learning approach that aims to fully automated pipeline for face detection and recognition of goats. Due to the high similarity and the lack of adequate dataset this problem is more complex than human face recognition. We composed two different publicly available datasets for detection and recognition. State-of-the-art convolutional neural networks (CNN) model are trained on this dataset. To evaluate the robustness of our approach, we compared it with different face recognition methods. The results show better performance with an accuracy of 96.4%. Furthermore, this paper reports 93%, 83%, 92% and 85% detection accuracy (average precision) for face, eye, nose and ear, respectively. The findings of this research could be helpful to improve animal health and welfare, individual monitoring, activity monitoring and phenotypic data collection. All the dataset and the related outcome are publicly available (https://doi.org/10.17632/4skwhnrscr.2). [ABSTRACT FROM AUTHOR]

• In this paper, we proposed a chrysanthemum image phenotype recognition framework based on transfer learning and bilinear Convolutional Neural Network for fine-grained chrysanthemum image data, we conducted extensive experiment and comparison, and experimental result proves that our proposal can achieve better performance and lower losses compared with other network models. • In addition, we highlight the on-line recognition, including training and testing. As one of the largest families of flowering plants, chrysanthemum has a wide variety and variation. There exists small phenotypic differences between classes, leading to the fact that classifying the class of chrysanthemum is a challenging fine-grained image classification problem. The fine-grained image classification task based on deep learning often concerns a large number of parameters and requires a long training time. For tackling these issues, this paper proposes a chrysanthemum image phenotype classification framework based on transfer learning and bilinear Convolutional Neural Network. After pre-processing the images, the symmetric VGG16 network is adopted as a feature extractor. The pre-trained parameters are then transferred to the proposed framework, which is divided into two stages for training the full connection layer and fine tuning all layers respectively. The phenotypic features of chrysanthemum output from the two networks are transposed and multiplied. Last the global features are input into the classification layer for classification. In this paper, a total of eight methods, including other bilinear network models, non-transfer learning and non-bilinear models, are compared with our proposal. The experimental result proves that our proposal can achieve better performance and lower loss, compared with other network models, by reaching at an accuracy rate of 0.9815 and a recall rate of 0.9800. [ABSTRACT FROM AUTHOR]

• Mark-Spectra can overcome the spatial correlations among adjacent wavelengths. • Mark-Spectra allows flexible controllability of characteristic wavelengths for convolutional operations. • Mark-Spectra introduces characteristics wavelengths labeling methods into deep learning-based spectral analysis. Spectral analysis is one of the most important and widely used methods for chemometrics in the field of agriculture, and convolutional neural network (CNN) models have achieved excellent performance on spectral analysis. The critical drawback of the CNN approach is that it preserves the spatial relationships among adjacent wavelengths, which contribute to collinearity and redundancies rather than relevant effective information. To confirm this observation, the distribution of characteristic wavelengths extracted by different methods (include F-test, importance weights, and CNN) are visualized in this paper. A convolutional neural network for quantitative spectral analysis, named Mark-Spectra, is presented to overcome spatial relationships and to improve the model performance. A layer (Mark layer) is introduced as part of Mark-Spectra, which is used to overcome spatial relationship of raw spectral data. Mark-Spectra model is compared with three CNN models using three open accessed visible and near infrared spectroscopic datasets (corn, wheat and soil). Mark-Spectra model outperforms the other three convolutional neural network models on two datasets (except dataset of wheat, due to lesser number of features), and it cost much less training time than the others. In addition, this paper compares Mark-Spectra with two classical neural network-based algorithms, principal component analysis – artificial neural network (PCA-ANN) and extreme learning machine (ELM). Mark-Spectra performed best in soil dataset, and ELM performed best in corn and wheat datasets, respectively. These results can illustrate that Mark-Spectra is still limited with the characteristic of raw spectral data (e.g., the number of samples and features), which is a fundamental fact of deep learning-based methods, but it performed better than the other CNN models and reduced the dependence of sample size due to overcoming spatial relationships. [ABSTRACT FROM AUTHOR]

• This paper reviews the latest research progress of image fusion technology in agriculture. • A variety of image fusion methods are introduced to improve the fusion image quality and monitoring accuracy. • The advantages and limitations of various image fusion techniques in crop and livestock monitoring were discussed. • The challenges and prospects of image fusion technology in agriculture are described. Crop and livestock monitoring are essential for guiding agricultural production and improving agricultural yield by facilitating the early prevention and treatment of diseases, yield prediction, and automatic harvesting. However, traditional monitoring methods have some shortcomings, including by humans and through a single sensor, manual monitoring method is time-consuming and labor-intensive, monitoring methods using a single sensor are inaccurate. Therefore, research on improving the methods of crop and livestock monitoring is essential. Image fusion technology can help provide important information from two or more images that can be used to obtain comprehensive agricultural information. After discussing the advantages and disadvantages of various of image fusion methods in agriculture, this paper reviews the application of image fusion in crop recognition and detection, planting area estimation, plant diseases and pest detection, and livestock health assessment and classification. It also highlights the main challenges to the successful application of image fusion technology to agriculture, and discusses directions of future research and development directions in the area. [ABSTRACT FROM AUTHOR]

• Proposing a top-down keypoint detection method to assist pruning bearing branch while picking fruit. • Generating candidate area by extending the citrus area to the citrus stem direction. • Fusing multi-level features from the same and different feature scales to extract information of bearing branch keypoint in the candidate area. • Achieving accurate detection of bearing branch keypoint with a smaller model size and lower computing cost. Automated orchard operation has been a firm goal of fruit farmers for a long time. Deep learning-based approaches have been widely used to improve the performance of fruit detection, branch pruning, production estimating and other agricultural operations. This paper proposes a novel method to detect keypoint on the branch, which enables branch pruning during fruit picking. Specifically, a top-down framework for bearing branch keypoint detection is developed. First, a candidate area is generated according to the fruit-growing position and the fruit stem keypoint detection, which provides an attention region for further keypoint detection. Second, a multi-level feature fusion network which combines features in the same spatial sizes (intra-level) and from different spatial sizes (inter-level) is proposed to detect keypoint within the candidate area. The network can learn the spatial and semantic information and model the relationship among bearing branch keypoints. In addition, this paper constructs a citrus bearing branch dataset, which contributes to comprehensively evaluating the proposed method. Evaluation metrics on the dataset indicate the proposed method reaches an AP of 77.4% and an accuracy score of 84.7% with smaller model size and lower computing power consumption, which significantly outperforms several state-of-the-art keypoint detection methods. This study provides the possibility and foundation for performing automatic branch pruning during fruit harvesting. [ABSTRACT FROM AUTHOR]

• Propose a more concise method for independent identification of crops and diseases. • A trilinear CNN model was constructed for crop and disease identification. • Improve the generalization of new models using pre-trained models on specific data. Plant diseases may threaten the safety of crops around the world, and timely detection of crop diseases and accurate determination of disease species are important to protect crop safety and control the spread of diseases. Recent studies have proposed the application of modern automatic recognition systems based on convolutional neural networks to the identification tasks of multiple crops and diseases. Although some research results have been achieved, studies have shown that these models are not optimal because they are susceptible to features unrelated to crop diseases, and have poor application ability in real-world environments. Therefore, this paper proposes a more concise method that separating the crop and disease identification and classify them independently, and demonstrates that it is more effective than the traditional crop-disease pairs approach. Meanwhile, we constructed a trilinear convolutional neural networks model using bilinear pooling and used images obtained in a real-world environment for the study of crop disease identification. The crop and disease identification accuracies achieved 99.99% and 99.7% on the test set in a controlled laboratory environment, and 84.11% and 75.58% on the test set in a real-world environment, respectively. The work in this paper improves the application value of crop disease identification research in the real-world. [ABSTRACT FROM AUTHOR]

• A rapid, non-invasive, and environmentally friendly solution is presented to identify 50 quality grades of 21 different tea brands of black tea and green tea. • TeaNet, TeaResnet, and TeaMobilenet are firstly proposed to process NIR data and the spatial information of the transformed pseudo image could be exploited by TeaNet, TeaResnet, and TeaMobilenet • The accuracy of the proposed TeaResnet could reach up to 100%. • SNV is the best preprocessing method for NIR data among MC, SG, SNV, Norm and there is a positive correlation between standard deviation and model performance. A high standard of tea quality control is extremely important for the sake of the health protection of consumers. With the fast-growing of Deep Learning, the intelligent and labor-free authentication of the tea quality, becomes a substantially demanded yet unexplored area, with two key problems yet to be solved: (1) sampling of tea data, with a fast and environmentally friendly approach, (2) distinguishment of tea, in terms of building an accurate tea classifier. To solve the above two problems, NIR spectroscopy is used to collect the tea NIR data in a fast, non-invasive, and environmentally friendly manner. And this paper proposed a series of brand new convolutional neural networks(CNNs) for NIR-based tea data, called TeaNet, TeaResnet, and TeaMobilenet. Due to the success of the different network architectures in the classification task, this paper applies the basic convolution block, the residual block, and the inverted residual block into the network architecture designs, corresponding to TeaNet, TeaResnet, and TeaMobilenet. To serve as the input of networks, NIR data is transformed into a pseudo image and then the model exploits the information in NIR data to make a classification. In an extensive experimental study, we compare TeaNet, TeaResnet, and TeaMobilenet with the traditional machine learning algorithms like Support Vector Machine (SVM) and Random Forest. The experimental results indicate that TeaNet, TeaResnet, and TeaMobilenet could break through the accuracy bottleneck of traditional machine learning algorithms and reach up to a 100% accuracy rate. At the same time, the comparative experiments of the four preprocessing methods-Mean Centering (MC), Savitzky-Golay smoothing (SG), Standard Normal Variate (SNV), and Normalization(Norm) indicate that there is a positive correlation between standard deviation and classification performance. SNV is the best preprocessing method for tea near-infrared spectroscopy data with the greatest standard deviation. [ABSTRACT FROM AUTHOR]

• A method for estimating citrus 3D pose via a single RGB image is developed. • An image annotation tool to label the fruit 3D pose in a 2D image is designed. • The fruit navel point is detected and the direction of fruit rotation is predicted. • The demonstrated performance shows that it is helpful for non-destructive harvesting. Automated fruit harvesting is promising research in the development of agricultural modernization. However, the complex and non-structural orchard environment is extremely challenging. In order to meet the needs of different end-effectors and to improve the success rate of automatic fruit harvesting, it is critical to perform fruit pose estimation before picking operations. In this study, a citrus pose estimation method through a single RGB image is introduced. The rotation of the citrus pose is defined as a vector that passes through the center of the fruit, which is perpendicular to the plane where the fruit navel point is located. Simply speaking, a multi-task learning model named FPENet is proposed to simultaneously locate the fruit navel point and predict the fruit rotation vector. And a hyperparameter is introduced in the loss function to achieve the simultaneous convergence of multiple tasks. In addition, this paper designs a 2D image annotation tool and constructs a citrus pose dataset, which contributes to model training and also the algorithm evaluation. In the experiment, we evaluate and analyze each module of the proposed network structure, and verify its performance on a harvesting robot. The experimental results show that the FPENet achieves an 88.92 AP score on fruit navel point detection, and 11.13 ° on the average error of the rotation vector. Over 90% of rotation vectors have an angular error of less than 22.5 °. The harvesting success rate is 79.79%. This study offers a new idea for fruit pose estimation and provides the possibility and foundation for estimating fruit pose with a 2D image input. [ABSTRACT FROM AUTHOR]

Tree-ring analysis is widely used in many different fields of science. Among others, information carried by annual tree rings allows determining the rates of environmental changes and the timing of events. The analysis of the tree rings requires prior detection of tree-ring boundaries which is traditionally performed manually with the use of stereoscope, a moving table, and a data recorder. This is, however, time-consuming and very cumbersome, especially in the case of long tree-ring series. Several approaches to an automatic detection of tree-ring boundaries exist; however, they use basic image processing techniques. As a result, their accuracy is limited, and their application is restricted mainly to conifer wood where the tree-ring boundaries are clearly defined. There also exists some commercial software, however, none of them is perfect as they fail when applied to the ring-porous wood type. Therefore this paper proposes a DeepDendro approach i.e., an automatic tree-ring boundary detector built upon the U-Net convolutional network. To the authors’ best knowledge this is the first study which applies ConvNets for an automatic detection of tree rings. The performance of the existing approach was tested on the dataset of images of wood cores of three species that represent the ring-porous type of the anatomical structure ( Quercus sp., Fraxinus excelsior L., and Ulmus sp.). The testing dataset contained over 2500 of tree-ring boundaries, 96% of which were determined correctly by the proposed method. The corresponding precision is at the level of 0.97 which confirms that only a few false boundaries were introduced by the DeepDendro approach. The results were obtained automatically without any user interaction. [ABSTRACT FROM AUTHOR]

• Chicken sounds were intercepted from complex noise. • A new method CSCNN was proposed in this paper. • CSCNN used to detect avian influenza-infected chickens via chicken sound. • The recognition accuracy is high to 97.43%. The modern poultry industry is large-scale and breeding-intensive, making the spread of disease in poultry easier, faster and more harmful. Avian influenza (AI) is the most important disease in poultry, and prevention and detection of avian influenza in poultry is a focus of scientific research and the poultry industry. In this paper, a new sound recognition method, the chicken sound convolutional neural network (CSCNN), is proposed for detection of chickens with avian influenza. According to the spectral differences in environmental noise, chicken behaviour noise and chicken sound, a method was designed to extract the chicken sound from complex sound data. Four features of the chicken sounds were calculated and combined into feature maps, including Logfbank, Mel Frequency Cepstrum Coefficient (MFCC), MFCC Delta and MFCC Delta-Delta. Finally, the sounds of healthy chickens and chickens with avian influenza were recognized using CSCNN. In the experiment, the recognition accuracies of CSCNN via spectrogram (CSCNN-S) were 93.01%, 95.05%, and 97.43% on the 2nd, 4th, and 6th day after injection with the H9N2 virus, and the recognition accuracies of CSCNN with feature mapping (CSCNN-F) were 89.79%, 93.56%, and 95.84%, respectively. The experimental results show that the method proposed in this paper can be used to quickly and effectively detect avian influenza-infected chickens via chicken sound. [ABSTRACT FROM AUTHOR]

• An automatic image segmentation algorithm is proposed to remove the background information of the image and improve the disease recognition accuracy in practical applications. • A more representative dataset of crop disease leaves is constructed to train and test the deep learning model. • A cash crop disease identification system on mobile devices is designed with the correct recognition rate of more than 80% for 27 diseases of 6 crops. Using deep learning methods to identify cash crop diseases has become a current hotspot in the field of plant disease identification. However, recent studies have demonstrated that the complex background information of crop images from practical application and insufficient training data can cause the wrong recognition of deep learning. To address this problem, in this paper we present an identification method of cash crop diseases using automatic image segmentation and deep learning with expanded dataset. An Automatic Image Segmentation Algorithm(AISA) based on the GrabCut algorithm is designed to remove the background information of images automatically while retaining the disease spots. It doesn't need to select the object manually during image processing and is of much lower time cost compared with the GrabCut algorithm. The MobileNet Convolutional Neural Network(CNN) model is selected as the deep learning model and plenty of crop images from the Internet and practical planting bases are added to expand the public dataset PlantVillage for the purpose of improving the generalization ability of MobileNet. The images are processed by the AISA before they can be used for extracting disease features, which reduces calculations significantly and ensures that the disease features of the crop leaf can be extracted accurately. Moreover, we design a cash crop disease identification system for mobile smart devices. The experimental results show that the system has a correct recognition rate of more than 80% for the 27 diseases of 6 crops described in this paper and then has a high value of practical application. [ABSTRACT FROM AUTHOR]

• We propose a novel pest detection pipeline towards large-scale multiple species of pest data. • A new Contextual information fusion mechanism (CAN) is integrated into DeepPest for pest detection. • A novel pest detection model (MDM) is proposed to address the pest feature vanishing problem in pest detection models. Automatic in-field pest detection and recognition using mobile vision technique is a hot topic in modern intelligent agriculture, but suffers from serious challenges including complexity of wild environment, detection of tiny size pest and classification of multiple classes of pests. While recent deep learning based mobile vision techniques have shown some success in overcoming above issues, one key problem is that towards large-scale multiple species of pest data, imbalanced classes significantly reduce their detection and recognition accuracy. In this paper, we propose a novel two-stages mobile vision based cascading pest detection approach (DeepPest) towards large-scale multiple species of pest data. This approach firstly extracts multi-scale contextual information of the images as prior knowledge to build up a context-aware attention network for initial classification of pest images into crop categories. Then, a multi-projection pest detection model (MDM) is proposed and trained by crop-related pest images. The role of MDM can combine pest contextual information from low-level convolutional layers with these in high-level convolutional layers for generating the super-resolved feature. Finally, we utilize the attention mechanism and data augmentation to improve the effectiveness of in-field pest detection. We evaluate our method on our newly established large-scale dataset In-Field Pest in Food Crop (IPFC) and sufficient experimental results show that DeepPest proposed in this paper outperforms state-of-the-art object detection methods in detecting in-field pest. [ABSTRACT FROM AUTHOR]

Monitoring the number of insect pests is a crucial component in pheromone-based pest management systems. In this paper, we propose an automatic detection pipeline based on deep learning for identifying and counting pests in images taken inside field traps. Applied to a commercial codling moth dataset, our method shows promising performance both qualitatively and quantitatively. Compared to previous attempts at pest detection, our approach uses no pest-specific engineering which enables it to adapt to other species and environments with minimal human effort. It is amenable to implementation on parallel hardware and therefore capable of deployment in settings where real-time performance is required. [ABSTRACT FROM AUTHOR]

• An RGB-D camera is mounted on the UAV to collect RGB and depth images. • Combine object detection with semantic segmentation. • A method combining CNN and RGB-D image information. • The 3D localization of branch picking points based on deep learning. • The performance of the whole scheme was evaluated by orchard experiments. Longan is a famous speciality fruit and cultivated medicinal plant that has important edible and medicinal value; how to improve productivity in harvest is an important issue. At present, longan is mainly planted in hilly areas. For complex site conditions and tall trees, the ground harvesting machineries cannot work normally. In this study, aiming at harvesting longan fruit using unmanned aerial vehicles, a method combining an improved YOLOv5s, improved DeepLabv3+ model and depth image information is proposed, which is used for the three-dimensional (3D) positioning of branch picking points in complex natural environments. First, the improved YOLOv5s model is used to quickly detect longan fruit skewers and the main fruit branches from a complex orchard environment. The correct main fruit branch is obtained according to its relative position relationship and is extracted as the input to the semantic segmentation model. Second, using the improved DeepLabv3+ model, the image extracted in the previous step is semantically segmented to obtain the 2D coordinate information of the main longan fruit branches. Finally, combined with the growth characteristics of a longan fruit string, RGB-D information fusion is carried out on the main fruit branches in 3D space to obtain the central axis and pose information of the main fruit branches, and the 3D coordinates of the picking points are calculated, which provides destination information for a longan harvesting drone. To verify the effectiveness of the proposed method, an experiment for identifying and locating the main fruit branches and picking points was carried out in a longan orchard. The experimental results show that the longan string fruit and main fruit branch detection accuracy is 85.50%, and the main fruit branch semantic segmentation accuracy is 94.52%. The whole algorithm takes 0.58 s in the actual scene and can quickly and accurately locate the picking points. In summary, this paper fully exploits the advantages of the combination of a convolutional neural network and RGB-D image information, further improving the efficiency of longan harvesting drones in accurately positioning picking points in 3D space. [ABSTRACT FROM AUTHOR]

• Propose a transfer-learning based model that captures the semantics of diseased plant leaf. • Fusion of Deep features and HOG feature for effective capture of rich local spatial properties in plant leaf images. • Introduce the centre-loss constraint to enhance the discriminative ability of the fused features. With the continuing changes in the structure of plant and cultivation patterns, new diseases are constantly appearing on the leaves of plant, exacerbating the threat to food security and agricultural production in many areas of the world. Thus, a rapid and accurate recognition of various diseases in plant will not only significantly reduce unnecessary planting costs, but also alleviate the economic losses and environmental pollution caused by incorrect disease diagnosis. Recent advances in deep learning have improved the performance in recognizing plant leaf diseases. In this paper, we present a general framework for recognizing plant diseases. Firstly, we propose a deep feature descriptor based on transfer learning to obtain a high-level latent feature representation. Then, we integrate the deep features with traditional handcrafted features by feature fusion to capture the local texture information in plant leaf images. In addition, centre loss is incorporated to further enhance the discriminative ability of the fused feature. The centre loss simultaneously minimizes intra-class distance and maximizes inter-class distance to learn both compact and separate features. Extensive experiments have been conducted on three publicly available datasets (two Apple Leaf datasets and one Coffee Leaf dataset) to validate the effectiveness of proposed method. The propose method achieves 99.79%, 92.59% and 97.12% classification accuracies on the three datasets, respectively. The experiment results demonstrate that the proposed method effectively captures the discriminative feature representation for distinguishing plant leaf diseases. [ABSTRACT FROM AUTHOR]

• Describing overall features by capturing apple images from multiple perspectives. • Lightweight CNN helps model more practical. • Aggregating spatial apple features is realized by bilateral long short-term memory. • The performance and efficiency are far excellent with a 99.23% accuracy. Apple grading is one of the typical tasks in food grading. For the image-based apple grading task, the variety of apple shapes and the incompleteness of single-view visual information pose great difficulties in achieving high accuracy of apple grading. This paper proposes a novel multi-view spatial network to tackle the apple grading task, which incorporates apple size information as one of the grading criteria, to solve those problems that affect the accuracy of grading. Specifically, we first use well-pretrained lightweight CNNs to extract low-level features of each view of apple. Second, we formulate a spatial feature aggregation module through a bidirectional LSTM and mean-pooling to exploit the correlative information of apple from multiple views. Besides, we elaborately design a multi-view apple data collection scheme for apple grading dataset construction. The experimental results indicate that the accuracy of Multi-View Spatial Network reaches 99.23%, which is quite promising in apple grading task and outperforms some state-of-the-art grading methods. [ABSTRACT FROM AUTHOR]

• An adaptive feature fusion pyramid network is developed to extract the optimal features. • A feature augmentation module is introduced for reducing the information loss of pest features. • A multi-stage detector is introduced for recognizing and classifying multi-classes pests. The accurate and robust crop pest detection system is an important step to enable the reliable forecasting of agricultural pest in the community of precision agriculture, attracting great attention in many countries. For achieving the automatic recognition and detection of agricultural pest, previous methods adopt image processing-based methods, leading to lower efficiency. Then, machine vision-based methods are introduced into crop pest detection by using hand-crafted feature descriptors, improving the detection precision and speed. However, the manual feature is powerless for precise recognition. Considering powerful ability of feature extraction of convolutional neural network(CNN), we have developed a CNN-based method for multi-classes pest detection under complex scenes. In this paper, an adaptive feature fusion is introduced into feature pyramid network for extracting richer pest features. Then, an adaptive augmentation module has been developed for reducing the information loss of the highest-level feature maps. Finally, a two-stage region-based convolutional neural network (R-CNN) was built for refining predicted bounding boxes, which can obtain the categories and locations of pests of each image. We have conducted large quantities of comparison experiments on AgriPest21 dataset. Our method could achieve an accuracy of 77.0%, which significantly outperforms other state-of-the-art methods, including SSD, RetinaNet, FPN, Dynamic R-CNN, and Cascade R-CNN. [ABSTRACT FROM AUTHOR]

• Identification and segmentation of weed species is of great importance. • Semantic segmentation on natural images requires enormous manual annotation. • The proposed architecture increases the performance in distinguishing between weed species. • The proposed methodology reduces the annotation effort to one third. • The use of synthetic datasets doubles the segmentation performance. Weeds compete with productive crops for soil, nutrients and sunlight and are therefore a major contributor to crop yield loss, which is why safer and more effective herbicide products are continually being developed. Digital evaluation tools to automate and homogenize field measurements are of vital importance to accelerate their development. However, the development of these tools requires the generation of semantic segmentation datasets, which is a complex, time-consuming and not easily affordable task. In this paper, we present a deep learning segmentation model that is able to distinguish between different plant species at the pixel level. First, we have generated three extensive datasets targeting one crop species (Zea mays), three grass species (Setaria verticillata, Digitaria sanguinalis, Echinochloa crus-galli) and three broadleaf species (Abutilon theophrasti, Chenopodium albums, Amaranthus retroflexus). The first dataset consists of real field images that were manually annotated. The second dataset is composed of images of plots where only one species is present at a time and the third type of dataset was synthetically generated from images of individual plants mimicking the distribution of real field images. Second, we have proposed a semantic segmentation architecture by extending a PSPNet architecture with an auxiliary classification loss to aid model convergence. Our results show that the network performance increases when supplementing the real field image dataset with the other types of datasets without increasing the manual annotation effort. More specifically, the use of the real field dataset obtains a Dice-Søensen Coefficient (DSC) score of 25.32. This performance increases when this dataset is combined with the single-species class dataset (DSC= 47.97) or the synthetic dataset (DSC= 45.20). As for the proposed model, the ablation method shows that by removing the proposed auxiliary classification loss, the segmentation performance decreases (DSC= 45.96) compared to the proposed architecture method (DSC= 47.97). The proposed method shows better performance than the current state of the art. In addition, the use of proposed single-species or synthetic datasets can double the performance of the algorithm than when using real datasets without additional manual annotation effort. [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 new disease evaluation method fusing deep learning and fuzzy logic is proposed. • DeepLabV3+ with ResNet50 backbone is utilized for images segmentation. • A fuzzy reference system is introduced for grape disease grades classification. • Two complementary indices are given to determine the disease grades. Grape black measles disease may be one of the best known, longest studied and most destructive of all plant diseases, which ultimately reduces productivity and quality of products. Timely, effective and accurate evaluation of grape black measles disease is acknowledged as a crucial step in field management. In this paper, an effective automatic detection and severity analysis method is proposed for grape black measles disease based on deep learning and fuzzy logic. In the first phase, the state-of-the-art ResNet50-based DeepLabV3+ semantic segmentation model is trained for pixel-level predictions in images of individual leaves exhibiting pathological lesions caused by fungus. In this way, the extracted features including ROI and POI are obtained. Second, for each feature, the fuzzy rule-based system is developed for predicting the harm degree of disease. Also, appropriate membership functions for the inputs and output are considered for fuzzification and defuzzification purposes in the fuzzy logic system. Finally, grape leaves are divided into grades of Healthy, Mild, Medium and Severe. Experimental results report an overall classification accuracy of 97.75% on the hold-out test dataset. It is concluded that the DeepLabV3+ model-based fuzzy reference system can be used effectively to classify grape leaves with different disease risks, based on combination of image analysis and statistical calculation. [ABSTRACT FROM AUTHOR]

• Proposed DTL-SE-ResNet50 for vegetable diseases identification. • Our model integrated the ResNet50 model, SENet, and dual transfer learning. • Our model provided the superior vegetable diseases identification ability. Artificial Intelligence for disease identification is currently the focus of great research interest. Nonetheless, the approach has some problems, for example, identification takes a long time, has low accuracy, and is often limited to a single disease type. Here, we aimed to identify tomato powdery mildew, leaf mold and cucumber downy mildew against simple and complex backgrounds. We developed a vegetable disease identification model, DTL-SE-ResNet50, optimized by SENet and pre-trained by ImageNet to form a new model, SE-ResNet50. The SE-ResNet50 model was trained with the AI Challenger 2018 public database to obtain a new weight. The SE-ResNet50 model with the new weight was then trained by dual transfer learning with a self-built database to create the DTL-SE-ResNet50 model for the identification of vegetable diseases. The model was compared with convolutional neural networks EfficientNet, AlexNet, VGG19, and Inception V3. The experimental results showed that with the same experimental conditions, the identification precision of the new model reached 97.24%, and processing of a single image required 0.13 s. Compared with DTL-CBAM-ResNet50 and DTL-SA-ResNet50, three models has almost the same precision, but time consumption of DTL-SE-ResNet50 was 0.02 s higher than that of DTL-CBAM-ResNet50. Although the time consumption of DTL-SA-ResNet50 was 0.02 s higher than the proposed model, the precision was lower. At the same time, compared with the dual transfer learning model, the model's precision was 4.1% higher, and the processing of a single image was 0.06 s shorter. Compared with convolutional neural networks, the precision of DTL-SE-ResNet50 was 3.19% higher than the best result, the time consumption of a single image was 0.58 s shorter; Recall and F1 also increased. The method proposed in this paper has high identification precision and short identification time, and it meets the requirements for accurate and rapid identification of vegetable diseases. [ABSTRACT FROM AUTHOR]

• Detect and locate fruits based on deep learning is the key to the success of fruit picking. • UAV fuzzy image preprocessing is the basic condition for longan detection. • Reduce the number of calculations and parameters is vital for target detection task. • The positioning strategy of suitable cutting points on fruit branches is elaborated in detail. • The performance of picking platform is evaluated by performing picking experiments. In agriculture, fruit picking robots on the ground have difficulty adapting to the terrain conditions of mountain orchards and cannot pick longan fruit from tall longan trees. In this paper, aiming to allow picking of longan fruit by unmanned aerial vehicles (UAVs), a deep learning-based scheme to quickly and accurately detect and locate suitable picking points on fruit branches is proposed. The scheme includes a UAV fuzzy image preprocessing method, longan detection based on a convolutional neural network (CNN), red, green, blue and depth (RGB-D) information fusion and an accurate target location strategy. First, the UAV is equipped with an Intel Realsense D455 camera, which collects longan images from the front for training and testing the model. Second, the lightweight MobileNet backbone network is used to improve the performance of the You Only Look Once version 4 (YOLOv4) model in feature extraction. The results for the test set show that compared with the classical feature pyramid network (FPN), YOLOv3 and YOLOv4 models, this model reduces the computation, parameters and detection time of the model. Compared with MobileNet single-shot multibox detector (MobileNet-SSD) and YOLOv4-tiny, this model exhibits improved detection accuracy. Third, according to the target detection result map, a strategy is formulated to accurately determine the suitable picking point on the main branch of the result. Finally, the performance of the improved model and picking platform in the harvest scene is evaluated by performing picking experiments in a longan orchard. In summary, we fully exploit the advantages of the combination of UAVs, RGB-D cameras and CNNs to improve the speed and accuracy of target detection and location for longan picking by UAVs based on vision. [ABSTRACT FROM AUTHOR]

• A deep convolutional neural network for the detection of Solanum rostratum Dunal. • A method for UAV image cropping to improve the accuracy and speed of weed pixel segmentation. • A method for calculating coverage that can be used to assess the extent of invasive weed damage. • Analyzed the effect of UAV flight height for weeds detection. Solanum rostratum Dunal is a common invasive weed that can cause significant harm to local natural environments and ecosystems. The prerequisite for preventing and managing the invasion of Solanum rostratum Dunal is timely detection and reasonable assessment of its damage level. Therefore, this paper proposes a deep learning-based image segmentation method for the detection of the invasion degree of Solanum rostratum Dunal. The Solanum rostratum Dunal images are acquired by UAV and then cropped into sub-images of the same size following a specific processing method. The sub-images are fed into a U-Net based convolutional neural network DeepSolanum-Net for processing. The pixels belonging to the Solanum rostratum Dunal plants are extracted and labeled after the sub-images pass through the DeepSolanum-Net. All the processed sub-images are stitched and reduced to the size of the original image, and all the target pixels belonging to the Solanum rostratum Dunal in the original image are segmented out from the background image. The coverage rate and the area covered by the Solanum rostratum Dunal on the ground are calculated based on the image segmentation results and the flight altitude of the UAV. A field test is executed and the test results demonstrate that the recognition precision of effective pixels of Solanum rostratum Dunal plants reach 89.95% and the recall rate reach 90.3% when the proposed method is used. [ABSTRACT FROM AUTHOR]

• A novel CNN-based pest monitoring system is designed and developed. • RPSN and SSM can improve the performance of pest monitoring. • Comprehensive experiments are provided for verifying the performance of our method. Due to cost-effectiveness and efficient automation, image analytic based automatic pest monitoring techniques are widely utilized in specialized control of pests in the agricultural crops industry. They could achieve good recognition performance in certain species of pests in site-specific environment, but are sensitive to content and characteristics of image like appearance variances and clustered background. For pests with small size and indistinct features like rice planthopper, it is hard to manually and timely select suitable features. In this paper, we propose an effective CNN based automatic hand-held mobile pest monitoring system to detect and count rice planthoppers. A rice planthopper search network (RPSN) approach is proposed for automatically extracting multiple high-quality proposal regions from large-scale pest images with tiny objects. Additionally, sensitive score matrix (SSM) is employed to further enhance the performance of classification and bounding box regression. The experimental results under the proposed approaches evaluating three types of density pest images show that our system performs well on detecting rice planthoppers in non-specific wild environment with recognition recall up to 91% in industrial circumstance, which outweighs the state-of-the art approaches. [ABSTRACT FROM AUTHOR]

• Spaceborne one-shot hyperspectral imagery for crop mapping in cloudy season. • The three-dimensional CNN model is found to be competent for identifying crops. • The in-depth features are analyzed and visualized using the t-SNE technique. • Results show one-shot hyperspectral images is superior to multi-temporal images in cloudy season. Crop mapping is essential for agricultural management, economic development planning, and ecological conservation. Remote sensing with a large field of view provides us with a potential technique for large-scale crop mapping. However, most of the previous studies have focused on multi-temporal crop mapping, requiring multiple imaging over a period of time, which is impossible in the cloudy season due to the absence of clear atmospheric windows. Recently, with the progress of spaceborne hyperspectral imaging technology, wide-width hyperspectral satellite images, which provide abundant spectral and spatial structure information, have made the precision crop mapping of large areas possible. This paper focuses on deep learning based crop mapping using one-shot hyperspectral satellite imagery, where three convolutional neural network (CNN) models, i.e., 1D-CNN, 2D-CNN, and 3D-CNN models, are applied for end-to-end crop mapping. In addition, a manifold learning based visualization approach, i.e., t-distributed stochastic neighbor embedding (t-SNE), is introduced to illustrate the discriminative ability of the deep semantic features extracted by the different CNN models. To demonstrate the advantages of one-shot hyperspectral satellite images, an experiment was designed to compare the crop mapping performance of different remote sensing data sources, where both mono-temporal and multi-temporal multispectral images (MSIs) of the same research area were introduced for a systematic comparison. The classification accuracy when using hyperspectral satellite images was found to reach more than 94%, which was much better than that when using mono-temporal MSIs, and was comparable to the result when using multi-temporal MSIs. These findings will be important for the application of hyperspectral data when mapping large-area crop landscapes, and they confirm the potential of CNN models, particularly 3D-CNN models, for crop recognition. [ABSTRACT FROM AUTHOR]

[Display omitted] • A lightweight CNN model (LightNet) is presented to classify the quality of zizania. • A compressed block is designed to reduce the computation complexity. • Extensive experiments are conducted with the automatic grading device. Zizania, one of aquatic vegetables, needs to be graded before entering market for assuring the product quality. However, it is time-consuming, tedious, labor-intensive, inaccurate and expensive to assess qualitatively and grade zizanias manually. This paper gives an effective solution to automatically classify fresh zizania into two categories, high quality and defective quality, by using the deep learnt features from the appearances. A new architecture of convolutional neural network, called LightNet, has been proposed and described. Specifically, it is composed of many compressed blocks, which is designed to reduce the computation complexity mainly by converting serial down-sampling and convolution operation to parallel structure. We evaluate the proposed architecture on the zizania image dataset collected by ourselves and integrate the algorithm in the automatic grading device. The experimental results show that the accuracy rate achieves 95.62% and the speed of inference quality is around 47 ms per zizania image. The proposed LightNet for automate classification has less parameters and lower computation complexity than popular networks, while maintaining the comparable accuracy in the task of grading zizanias. It obtains 99.31% accuracy in the task of grading apples. The result proves that it can be extended to other tasks about classification. [ABSTRACT FROM AUTHOR]

• A fully automatic approach for tree species identification is proposed. • The method is dedicated to images of scanned wood cores. • The method employs a residual convolutional encoder network in a sliding window setup. • The method was tested on a dataset of images representing 14 European tree species. • Tree wood species of about 99% core samples were identified correctly. This paper tackles the problem of automatic tree species identification from scanned images of wood cores. A convolutional neural network with residual connections is proposed to perform this task. The model is applied to consecutive image patches following the sliding window strategy to recognize a patch central pixel's membership. It then decides about the resulting tree species via a majority voting. The model's performance was assessed concerning a dataset of 312 wood core images representing 14 European tree species, including both conifer and angiosperm (ring-porous and diffuse-porous) wood. Two tasks were considered, including wood patch classification and wood core classification. In these tasks, the proposed model correctly recognized species of almost 93% the wood image patches and 98.7% of wood core images. It also outperformed the state-of-the-art convolutional neural network-based competitor by 9% and 3%, respectively. The influence of the model's parameters and training set-up on its performance is analyzed in the manuscript to ensure the highest recognition rates of wood species. The source code of the proposed method is released together with the corresponding image dataset to facilitate the reproduction of results. [ABSTRACT FROM AUTHOR]

• A systematic review on the applicability of machine learning and image processing in pest management. • Review of 33 relevant research studies. • The use of traps, image datasets, or plant images were the most used techniques. • Great opportunity to exploit the insect overlapping problem, customizations for small objects, and instance segmentation. The monitoring of pests in the field or lab experiments allows to identify the variation of infection levels and to enhance the development of integrated pest management programs. The use of traps to capture insects is an alternative in different crops and regions. However, identification and manual counting of captured specimens is often time-consuming, requires taxonomic knowledge, and relies on the expertise of specialists. Therefore, the automation of this process could reduce cost, increase accuracy, and scalable the analysis. Current computer vision and artificial intelligence techniques can identify objects of interest in digital images in a timely and accurate manner. Hence, this paper presents a survey considering the following Computer Science digital research databases: ACM, IEEE, IET, DBLP, ScienceDirect, Scopus, SpringerLink, and Web of Science. We found three hundred studies, published between 2015 to 2019, of which thirty-three were selected based on the eligibility criteria. Results showed the use of convolutional neural network approaches, techniques to improve feature extraction, the lack of treatment to insect overlapping, and the non-use of instance segmentation via deep learning. [ABSTRACT FROM AUTHOR]

• A dense CNN model is presented for diseases recognition and classification in corn leaf. • A comprehensive step-by-step discussion of proposed model is given. • Rigorous experimental evaluation is presented. • Performance comparison is done with state-of-the-art models. An optimized dense convolutional neural network (CNN) architecture (DenseNet) for corn leaf disease recognition and classification is proposed in this paper. Corn is one of the most cultivated grain throughout the world. Corn crops are highly susceptible to certain leaf diseases such as corn common rust, corn gray leaf spot, and northern corn leaf blight are very common. Symptoms of these leaf diseases are not differentiable in their nascent stages. Hence, the current research presents a solution through deep learning so that crop health can be monitored and, it will lead to an increase in the quantity as well as the quality of crop production. The proposed optimized DenseNet model has achieved an accuracy of 98.06%. Besides, it uses significantly lesser parameters as compared to the various existing CNN such as EfficientNet, VGG19Net, NASNet, and Xception Net. The performance of the optimized DenseNet model has been contrasted with the current CNN architectures by considering two (time and accuracy) quality measures. This study indicates that the performance of the optimized DenseNet model is close to that of the established CNN architectures with far fewer parameters and computation time. [ABSTRACT FROM AUTHOR]

• A prioritization method for grasping stacked fruit clusters is proposed. • A convolutional neural network for small objects is constructed. • A coarse-to-fine cascaded faster region convolutional neural network is proposed. • An improved transfer-learning scheme is used. Determining the priority order for grasping stacked fruit clusters such as those of grape and lychee is a challenging problem in robotic manipulation based on RGB-D images. Grasping fruit cluster stalks at the top of a stack can improve robotic grasping success and reduce fruit damage. Hence, stalk location is an important factor affecting grasping priority. Because of the disordered arrangement of fruit clusters, stalks are not neatly arranged in horizontal planes, and this makes it difficult to determine grasping priority by stalk vertical location. Therefore, this paper proposes a prioritization method for grasping stacked fruit clusters by extracting and classifying "stalk depth sets" in RGB-D images. For extracting small stalks with multiple background elements in RGB-D images, a convolutional neural network (CNN) is constructed to reduce the variation in resolution across the feature map and to retain details and local features. To reduce background interference, a coarse-to-fine cascaded Faster R-CNN based on multi-scale feature maps is proposed. A transfer learning scheme is used to improve the network's learning capability for stalk depth sets. Using grape cluster experiments with a robotic sorting system, the proposed method was verified and compared with the existing LeNet-5, AlexNet, VGG16, and Faster R-CNN methods. Compared with existing networks, the average precisions of cascaded Faster R-CNN for extracting stalk depth sets and constructed CNN trained by improved transfer learning for grasping prioritization increased by more than 8.01% and 3.79%, respectively. The experimental results demonstrate that the proposed method can accurately extract and classify stalk depth sets and improve grasping prioritization for stacked fruit clusters. [ABSTRACT FROM AUTHOR]

• A maize water stress recognition dataset is constructed to do deep learning. • Local features and global context information are extracted from the CNN. • Dimensionality reduction improves recognition performance and efficiency. • Maize water stress in early stage is detected and quantified. Water stress significantly influences normal maize growth. Fast and effective maize water stress detection is of great help to monitor the plant status and provide scientific guidance for crop irrigation. Most of the methods are based on manual measurements of soil water content, or laboratory imaging techniques, such as hyperspectral and thermal images at plant level. With the collection of 656 original maize plant images under natural environment, a novel maize leaf image dataset with different water stress levels (well-watered, reduced-watered and drought-stressed) was constructed. This paper considers maize water status detection as a fine-grained classification problem using local leaf images. Inspired by deep learning, a convolutional neural network (CNN) is applied for the first time to maize water stress recognition. In the designed CNN architecture, feature maps from different convolutional layers are merged. Through visualization and importance analysis of the multi-scale feature maps, several specific feature maps are selected as learned features, which provide a strong discrimination ability. An SVM classifier is finally trained using the feature representation as inputs. Compared with existing techniques, the proposed method achieves the satisfying classification performance with an accuracy of 88.41%. This study also provides a quantitative measure of water stress degree using a regression model. Experimental results demonstrate that the learned features perform better than hand-crafted features to detect water stress and quantify stress severity. The proposed framework can be deployed in practical applications for a non-destructive, near real-time, and automatic monitoring of plant water status in fields. [ABSTRACT FROM AUTHOR]

• A deep learning method based on CNN and LSTM was used to recognise pig aggression. • This study directly process video episodes rather than individual frames. • CNN features were input into LSTM framework to extract spatial-temporal features. • The proposed method could recognise aggressive episodes with an accuracy of 97.2%. • A frame skipping approach can improve the accuracy to 98.4% and halve running time. Aggression is considered as a major animal welfare problem in commercial pig farming. The aim of this study is to develop a deep learning method based on convolutional neural network (CNN) and long short-term memory (LSTM) to recognise aggressive episodes of pigs. Compared to previous studies of pig behaviours based on deep learning, this study directly process video episodes rather than individual frames. In the experiment, nursery pigs (8/pen) were mixed for 3 days and then 8 h of video was recorded in each day. From these videos, 600 aggressive 2 s-episodes were manually selected and then augmented into 2400 episodes by using horizontal, vertical and diagonal mirroring. From the videos, 2400 non-aggressive 2 s-episodes were also manually selected. 80% of the data were randomly allocated as training set and the remaining 20% as validation set. Firstly, the CNN architecture VGG-16 was used to extract spatial features. These features were then input into LSTM framework to further extract temporal features. Through fully connected layer, the prediction function Softmax was finally used to determine if the current episode is aggression or non-aggression. Using the proposed method, aggressive episodes could be recognised with an accuracy of 97.2%. This result indicates that this method can be used to recognise aggressive episodes of pigs. Additionally, this paper further investigates the validity of this method under the conditions of skipping frames and reducing the episode length. The results show that a frame skipping approach whereby 30 fps is reduced into 15 fps within each 2 s-episode can improve the accuracy into 98.4% and halve the total running time. [ABSTRACT FROM AUTHOR]

• Challenging class data imbalance issue in deep-learning plant disease control. • Proposing a novel translator to generate disease images from healthy ones. • Augmenting tomato disease dataset with synthetic diseased images. • Significantly enhanced disease detection performance. Acquisition of data in task-specific applications of machine learning like plant disease recognition is a costly endeavor owing to the requirements of professional human diligence and time constraints. In this paper, we present a pipeline that uses GANs in an unsupervised image translation environment to improve learning with respect to the data distribution in a plant disease dataset, reducing the partiality introduced by acute class imbalance and hence shifting the classification decision boundary towards better performance. The empirical analysis of our method is demonstrated on a limited dataset of 2789 tomato plant disease images, highly corrupted with an imbalance in the 9 disease categories. First, we extend the state of the art for the GAN-based image-to-image translation method by enhancing the perceptual quality of the generated images and preserving the semantics. We introduce AR-GAN, where in addition to the adversarial loss, our synthetic image generator optimizes on Activation Reconstruction loss (ARL) function that optimizes feature activations against the natural image. We present visually more compelling synthetic images in comparison to most prominent existing models and evaluate the performance of our GAN framework in terms of various datasets and metrics. Second, we evaluate the performance of a baseline convolutional neural network classifier for improved recognition using the resulting synthetic samples to augment our training set and compare it with the classical data augmentation scheme. We observe a significant improvement in classification accuracy (+5.2%) using generated synthetic samples as compared to (+0.8%) increase using classic augmentation in an equal class distribution environment. [ABSTRACT FROM AUTHOR]