Search

Background: Machine vision-based precision weed management is a promising solution to substantially reduce herbicide input and weed control cost. The objective of this research was to compare two different deep learning-based approaches for detecting weeds in cabbage: (1) detecting weeds directly, and (2) detecting crops by generating the bounding boxes covering the crops and any green pixels outside the bounding boxes were deemed as weeds., Results: The precision, recall, F1-score, mAP 0.5 , mAP0 .5:0.95 of You Only Look Once (YOLO) v5 for detecting cabbage were 0.986, 0.979, 0.982, 0.995, and 0.851, respectively, while these metrics were 0.973, 0.985, 0.979, 0.993, and 0.906 for YOLOv8, respectively. However, none of these metrics exceeded 0.891 when detecting weeds. The reduced performances for directly detecting weeds could be attributed to the diverse weed species at varying densities and growth stages with different plant morphologies. A segmentation procedure demonstrated its effectiveness for extracting weeds outside the bounding boxes covering the crops, and thereby realizing effective indirect weed detection., Conclusion: The indirect weed detection approach demands less manpower as the need for constructing a large training dataset containing a variety of weed species is unnecessary. However, in a certain case, weeds are likely to remain undetected due to their growth in close proximity with crops and being situated within the predicted bounding boxes that encompass the crops. The models generated in this research can be used in conjunction with the machine vision subsystem of a smart sprayer or mechanical weeder. © 2024 Society of Chemical Industry., (© 2024 Society of Chemical Industry.)

Advancing spatially resolved transcriptomics (ST) technologies help biologists comprehensively understand organ function and tissue microenvironment. Accurate spatial domain identification is the foundation for delineating genome heterogeneity and cellular interaction. Motivated by this perspective, a graph deep learning (GDL) based spatial clustering approach is constructed in this paper. First, the deep graph infomax module embedded with residual gated graph convolutional neural network is leveraged to address the gene expression profiles and spatial positions in ST. Then, the Bayesian Gaussian mixture model is applied to handle the latent embeddings to generate spatial domains. Designed experiments certify that the presented method is superior to other state-of-the-art GDL-enabled techniques on multiple ST datasets. The codes and dataset used in this manuscript are summarized at https://github.com/narutoten520/SCGDL., (© The Author(s) 2023. Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oup.com.)

BACKGROUND: Accurate weed detection is a prerequisite for precise automatic precision herbicide application. Previous research has adopted the laborious and time‐consuming approach of manually labeling and processing large image data sets to develop deep neural networks for weed detection. This research introduces a novel semi‐supervised learning (SSL) approach for detecting weeds in turf. The performance of SSL was compared with that of ResNet50, a fully supervised learning (FSL) method, in detecting and differentiating sub‐images containing weeds from those containing only turfgrass. RESULTS: Compared with ResNet50, the evaluated SSL methods, Π‐model, Mean Teacher, and FixMatch, increased the classification accuracy by 2.8%, 0.7%, and 3.9%, respectively, when only 100 labeled images per class were utilized. FixMatch was the most efficient and reliable model, as it exhibited higher accuracy (≥0.9530) and F1 scores (≥0.951) with fewer labeled data (50 per class) in the validation and testing data sets than the other neural networks evaluated. CONCLUSION: These results reveal that the SSL deep neural networks are capable of being highly accurate while requiring fewer labeled training images, thus being more time‐ and labor‐efficient than the FSL method. © 2024 Society of Chemical Industry. [ABSTRACT FROM AUTHOR]

Motivation Spatial clustering is essential and challenging for spatial transcriptomics' data analysis to unravel tissue microenvironment and biological function. Graph neural networks are promising to address gene expression profiles and spatial location information in spatial transcriptomics to generate latent representations. However, choosing an appropriate graph deep learning module and graph neural network necessitates further exploration and investigation. Results In this article, we present GRAPHDeep to assemble a spatial clustering framework for heterogeneous spatial transcriptomics data. Through integrating 2 graph deep learning modules and 20 graph neural networks, the most appropriate combination is decided for each dataset. The constructed spatial clustering method is compared with state-of-the-art algorithms to demonstrate its effectiveness and superiority. The significant new findings include: (i) the number of genes or proteins of spatial omics data is quite crucial in spatial clustering algorithms; (ii) the variational graph autoencoder is more suitable for spatial clustering tasks than deep graph infomax module; (iii) UniMP, SAGE, SuperGAT, GATv2, GCN, and TAG are the recommended graph neural networks for spatial clustering tasks; and (iv) the used graph neural network in the existent spatial clustering frameworks is not the best candidate. This study could be regarded as desirable guidance for choosing an appropriate graph neural network for spatial clustering. Availability and implementation The source code of GRAPHDeep is available at https://github.com/narutoten520/GRAPHDeep. The studied spatial omics data are available at https://zenodo.org/record/8141084. [ABSTRACT FROM AUTHOR]

Precision spraying can significantly reduce herbicide input for turf weed management. A major challenge for autonomous precision herbicide spraying is to accurately and reliably detect weeds growing in turf. Deep convolutional neural networks (DCNNs), an important artificial intelligent tool, demonstrated extraordinary capability to learn complex features from images. The feasibility of using DCNNs, including various image classification or object detection neural networks, has been investigated to detect weeds growing in turf. Due to the high level of performance of weed detection, DCNNs are suitable for the ground-based detection and discrimination of weeds growing in turf. However, reliable weed detection may be subject to the influence of weeds (e.g., biotypes, species, densities, and growth stages) and turf factors (e.g., turf quality, mowing height, and dormancy vs. non-dormancy). The present review article summarizes the previous research findings using DCNNs as the machine vision decision system of smart sprayers for precision herbicide spraying, with the aim of providing insights into future research. [ABSTRACT FROM AUTHOR]

Real-time monitoring of seedling emergence is vital for vegetable crop management and yield estimation. Traditionally, crop seedling emergence monitoring relies on low-efficient and time-consuming manual counting. To address this issue, this research proposed an efficient, fast, and real-time cabbage seedling counting method (combining the improved YOLOv8n, tracking algorithm, and image processing) to accurately track cabbage seedlings in the field and implement counting with an unmanned aerial vehicle (UAV). The improved YOLOv8n replaced the C2f Block in the YOLO backbone with a Swin-conv block and incorporated ParNet attention modules in both the backbone and neck parts. This enhancement enables the YOLOv8n to surpass the base model's performance, achieving a mAP50–95 of 90.3 %, representing a 14.5 % improvement. The experiments demonstrated the superior capabilities of the counting method in terms of speed and accuracy. In field experiments, the proposed Tracking algorithms-Swin-conv blocks-ParNet attention-YOLOv8n (TSP-yolo) counting method demonstrated consistent and reliable accuracy in counting cabbage seedlings while demanding only one-seventh of the time needed compared to the manual counting method. In summary, based on TSP-yolo and implemented through an UAV, the developed seedling emergence counting method demonstrated an excellent capability of counting cabbage seedlings, resulting in significant savings in human resources for crop management. • This study effectively addressed the issue of crop seedling emergence monitoring. • The improved YOLOv8n showed superior performance for cabbage identification. • TSP-yolo-based real-time counting model achieved an impressive accuracy. • The proposed method can accurately monitor seedling densities or emergence rates. [ABSTRACT FROM AUTHOR]

Deep learning-based methods for weed detection and precise herbicide application are promising for reducing herbicide input and weed control costs. However, training the neural network to recognize weeds requires annotating a large number of training images, which is time-consuming and labor-intensive. In addition, in turf sod farms, sods need to be periodically harvested, leading to varying turfgrass and bare soil areas, which increase the complexity of weed detection. To solve this problem, this research explored semi-supervised learning (SSL) methods to train image classification neural networks. The experiments were conducted to compare the training results using different SSL strategies with 100 and 200 labeled images at three image sizes of 240 × 240, 360 × 360, or 480 × 480 pixels. The training dataset images mainly contained purple nutsedge (Cyperus rotundus L.) and green kyllinga (Kyllinga brevifolia) at the pre-flowering or seedhead stage. The F 1 score, precision, and recall were used to evaluate the performance of the trained neural networks. The results showed that the network based on the FixMatch SSL strategy trained with the input images of 240 × 240 pixels exhibited the highest F 1 score, reaching 98.1% when trained with 100 labeled images and 98.2% when trained with 200 labeled images. To summarize, these results suggest that SSL achieved a great training performance with a small number of annotations. FixMatch SSL was the most effective neural network training strategy evaluated. For the weed detection task, it was observed that neural networks trained using an input image size of 240 × 240 pixels exhibited superior performance compared to the networks trained with other image sizes. In addition, employing the SSL method with only 200 labeled images enhanced the performance of the neural network, surpassing that of fully supervised learning (FSL) approaches. • The SSL strategy proved to be effective for weed detection with a small amount of labeled data. • The SSL strategy employed in FixMatch outperformed FSL. • The optimal performance of the trained model was achieved when utilizing input images at a resolution of 240 × 240 pixels. • Varying turf coverage and bare soil area barely had any meaningful impact on the performance of SSL for weed detection. [ABSTRACT FROM AUTHOR]

[Display omitted] • YOLO series were evaluated for discrimination of alfalfa pollinating bee species. • YOLOv3 and -v5 outperformed YOLOv7 and -R regarding model precision, recall, F 1 score. • YOLOv3 and -v5 discriminated honeybee, bumblebee, and leafcutting bee with 100% accuracy. • Proposed method would help study pollinator behaviour and associated pollen flow in alfalfa. Identifying insect pollinators and their roles in mediating pollen flow is critical to understand the potential gene flow risks of insect pollination-dependent crop species, such as alfalfa. This study was conducted to evaluate and compare the feasibility of You Only Look Once (YOLO) version 3 (YOLOv3), YOLOv5, YOLOv7, and YOLO Representation (YOLOR) to discriminate the three most common alfalfa pollinating bee species, including honeybee, bumblebee, and leafcutting bee. The metrics comparison results showed YOLOv3 and YOLOv5 outperformed YOLOv7 and YOLOR regarding model precision, recall, F 1 score, and mAP50 values. YOLOv3 and YOLOv5 could successfully discriminate the three different bee species with an accuracy of almost 100% (99.9%, 99.8%, and 100% accuracy for honeybee, bumblebee, and leafcutting bee for the two models, respectively). Comparatively, YOLOv7 could discriminate honeybee with an accuracy of 95% but was more likely to mistakenly discriminate bumblebee and leafcutting bee due to the relatively lower discriminating accuracy (87.3% and 66.2%, respectively). While the values of determined parameters for YOLOR were lower than YOLOv3 and YOLOv5, the higher precision (0.99680) along with recall (0.98721), F 1 (0.99198), mAP50 (0.99323), and mAP50-100 (0.89076) values indicate that this model could be able to obtain a favorable performance in discriminating the three bee species. In summary, the proposed method in this study has the potential for identifying the alfalfa pollinating bee species, studying the bees' flower-visiting behaviors, evaluating the risks of insect-mediated pollen flow, and thus contributing to the management of genetically engineered (GE) alfalfa transgene flow. [ABSTRACT FROM AUTHOR]

Knowledge transfer is a promising concept to achieve real-time decision-making for autonomous vehicles. This paper constructs a transfer deep reinforcement learning (RL) framework to transform the driving tasks in the intersection environments. The driving missions at the unsignalized intersection are cast into a left turn, right turn, and running straight for automated vehicles. The goal of the autonomous ego vehicle (AEV) is to drive through the intersection situation efficiently and safely. This objective promotes the studied vehicle to increase its speed and avoid crashing other vehicles. The decision-making policy learned from one driving task is transferred through three transfer rules in another driving mission and evaluated. Simulation results reveal that the decision-making strategies related to similar tasks are transferable and have a high success rate. It indicates that the presented control framework could reduce time consumption and realize online implementation. Therefore, the transfer RL concept is helpful for establishing the real-time decision-making policy for autonomous vehicles. [ABSTRACT FROM AUTHOR]

Security perception systems based on 5G-V2X have become an indispensable part of smart city construction. However, the detection speed of traditional deep learning models is slow, and the low-latency characteristics of 5G networks cannot be fully utilized. In order to improve the safety perception ability based on 5G-V2X, increase the detection speed in vehicle perception. A vehicle perception model is proposed. First, an adaptive feature extraction method is adopted to enhance the expression of small-scale features and improve the feature extraction ability of small-scale targets. Then, by improving the feature fusion method, the shallow information is fused layer by layer to solve the problem of feature loss. Finally, the attention enhancement method is introduced to increase the center point prediction ability and solve the problem of target occlusion. The experimental results show that the UA-DETRAC data set has a good detection effect. Compared with the vehicle detection capability before the improvement, the detection accuracy and speed have been greatly improved, which effectively improves the security perception capability based on the 5G-V2X system, thereby promoting the construction of smart cities. [ABSTRACT FROM AUTHOR]

Advanced algorithms can promote the development of energy management strategies (EMSs) as a key technology in hybrid electric vehicles (HEVs). Reinforcement learning (RL) with distributed structure can significantly improve training efficiency in complex environments, and multi-threaded parallel computing provides a reliable algorithm basis for promoting adaptability. Dedicated to trying more efficient deep reinforcement learning (DRL) algorithms, this paper proposed a deep q-network (DQN)-based energy and emission management strategy (E&EMS) at first. Then, two distributed DRL algorithms, namely asynchronous advantage actor-critic (A3C) and distributed proximal policy optimization (DPPO), were adopted to propose EMSs, respectively. Finally, emission optimization was taken into account and then distributed DRL-based E&EMSs were proposed. Regarding dynamic programming (DP) as the optimal benchmark, simulation results show that three DRL-based control strategies can achieve near-optimal fuel economy and outstanding computational efficiency, and compared with DQN, two distributed DRL algorithms have improved the learning efficiency by four times. [ABSTRACT FROM AUTHOR]

Machine vision-based precision herbicide application in wheat (Triticum aestivum L.) can substantially reduce herbicide input. However, detecting newly emerged weeds in wheat fields remains a challenging task. Current deep learning-based weed detection methods require the annotation of a large amount of data, which is both time-consuming and labor-intensive. To address this issue, this research improved a semi-supervised learning (SSL) algorithm based on consistency regularization and pseudo-labeling, and incorporated an attention mechanism. Compared to fully supervised learning (FSL) algorithms, the proposed method increased the classification accuracy by 16.5%, 17.84%, and 19.67% on datasets with 200 × 200, 300 × 300, and 400 × 400 pixel images, respectively, when only 100 labeled data per class were used. Overall, the developed machine vision models using the proposed method achieved weed detection with high accuracy while requiring much fewer labeled training images, and thus is more time and labor-efficient compared to an FSL algorithm. • A semi-supervised learning method was used for detecting weeds in wheat. • The addition of attention mechanisms improved the performance of weed detection. • The semi-supervised learning method out-performed fully supervised learning. • 200×200 pixels out-performed larger image sizes for semi-supervised learning. [ABSTRACT FROM AUTHOR]

Precision herbicide application can substantially reduce herbicide input, thereby cutting chemical costs and minimizing adverse environmental impacts. A smart sprayer prototype was designed and developed for precision herbicide application in turf. This is the first study evaluating the performances of a precision sprayer for weed control in turf in field conditions. The objectives of this research were to 1) evaluate and compare the performances of the traditional broadcast application and a newly developed precision spraying technology for control of weeds in dormant bermudagrass turf, and 2) investigate the influence of weed coverage on the spray volume requirement when using the precision spraying technology developed here. DenseNet, GoogLeNet, and ResNet were evaluated for discriminating the grid cells containing weeds (spray) with the grid cells containing bermudagrass turf exclusively (nonspray). All three neural networks had an F 1 score above 0.989 in the validation datasets. ResNet outperformed DenseNet and GoogLeNet with the highest F 1 scores (≥0.992) in the testing datasets. Applying herbicide only to turf areas infested with weeds saved a significant amount of the herbicide, while achieving the same level of weed control compared to the broadcast application. The developed precision spraying technology performed well and effectively reduced the amount of herbicide input applied to the dormant bermudagrass turf, compared to the broadcast herbicide application. Overall, the smart sprayer prototype developed in this research can be used for precision weed control in dormant turf, although its design needs to be further optimized. • A custom software was developed to detect and locate weeds using image classification neural networks. • A smart sprayer prototype was designed and developed for precision herbicide application on dormant bermudagrass turf. • The developed precision spraying technology effectively reduced the amount of herbicide input applied to the turf. [ABSTRACT FROM AUTHOR]

Precision application of specific herbicides to susceptible weeds can significantly save herbicide. This is the first study evaluating the performances of precision sprayer for weed control in turf based on the herbicide weed control spectrum in field conditions. The results showed that EfficientNet-v2 and ResNet never fall below 0.992 for discriminating and detecting the grid cells encompassing weeds susceptible to ACCase-inhibiting and synthetic auxin herbicides. MCPA, a synthetic auxin herbicide, is used to evaluate the performance of the developed smart sprayer for precision control of broadleaf weeds in dormant bermudagrass turf. The developed smart sprayer prototype detected and sprayed every grid cell containing broadleaf weeds in field experiments. Compared to the broadcast application, precision spraying of MCPA provided the same level of control of broadleaf weeds. By 18 days after treatment (DAT), the nontreated control had 13 weeds no. m−2, while the plots that received broadcast and precision spraying had 0 and 1 broadleaf weed plant no. m−2, respectively. Precision herbicide application according to the herbicide weed control spectrum (HWCS) with the developed smart sprayer provided the same level of broadleaf weed control and could save more herbicides compared to an approach without discriminating weed species. Overall, these findings clearly indicated that the developed smart sprayer prototype could effectively detect, discriminate, and spray herbicides onto the grid cells containing target weeds based on the HWCS. • Discriminating weed species based on their susceptibility to herbicides allows targeted herbicide application. • A smart sprayer prototype was designed and developed for precision herbicide application on turf. • Precision herbicide application with the smart sprayer achieved weed control equivalent to broadcast application. [ABSTRACT FROM AUTHOR]

Diabetic retinopathy is one of the leading reasons that causes blindness. And the segmentation of hard exudates in color fundus images is crucial for early diagnosis of diabetic retinopathy, which is a difficult task due to its uncertainty in size, shape and contrast. Class-balanced cross entropy (CBCE) loss is the most popular objective function for image segmentation task to solve the class-unbalance problem. However, we show that background pixels tend to be misclassified to hard exudates in CBCE since the loss for a misclassified background pixels is much smaller than that for a misclassified hard exudate pixel, which is called loss-unbalance problem here. A top-k loss is proposed in this paper, which considers the cases of both class-unbalance and loss-unbalance by focusing more over the hard-to-classify pixels. Moreover, a fast version of the top-k loss, named bin loss, is implemented for efficiency, which reduces the time complexity from O (n log n) of top-k loss to O (n), where n is the number of background pixels. We evaluated the proposed bin loss over two public datasets for hard exudates segmentation task, including e-ophtha EX and IDRiD. Furthermore, three popular models for image segmentation, HED, DeepLab v2, and FCRN, were used to evaluate the versatility of bin loss. Extensive experiments show that each model with the proposed bin loss performs better than that with CBCE loss, which demonstrates bin loss is versatile so that it can be applied to different models for performance improvement. Specially, for DeepLab over e-ophtha EX, the F-score increases 5.2 percentage points, and the area under the SE-PPV curve (AUC) increases 10.6 percentage points. Moreover, the AUC increases more than 4 percentage points over IDRiD dataset for both DeepLab and FCRN. The source code of bin loss is available at: https://github.com/guomugong/bin_loss. [ABSTRACT FROM AUTHOR]

Aiming at promoting the intelligent development of control technology for new energy vehicles and showing the outstanding advantages of deep reinforcement learning (DRL), this paper trained a VGG16-based road recognition convolutional neural network (CNN) at first. Lots of high-definition images of five typical roads are collected by the racing game Dust Rally 2.0, including dry asphalt, wet asphalt, snow, dry cobblestone, and wet cobblestone. Then, a time-varying driving environment model was established, involving driving images, road slope, longitudinal speed, and the number of passengers. Finally, a stereoscopic control network suitable for nine-dimensional state space and three-dimensional action space was built, and for parallel hybrid electric vehicles (HEVs) with the P3 structure, a deep q-network (DQN)-based energy management strategy (EMS) achieving multi-objective control was proposed, including the fine-tuning strategy of the motor speed to maintain the optimal slip rate during braking, the engine power control strategy and the continuously variable transmission (CVT) gear ratio control strategy. Simulation results show under the influence of some factors such as tree shade and image compression, the road recognition network has the highest accuracy for snow roads and wet asphalt roads. Three types of control strategies learned simultaneously by the stereoscopic control network not only maintain the near-optimal slip rate in the braking period but also achieve a fuel consumption of 4788.93 g, while dynamic programming (DP)-based EMS gets a fuel consumption of 4295.61 g. Moreover, even DP-based EMS only contains three states and two actions, the time consumed for DP-based EMS and DQN-based EMS to run the speed cycle of 3602s is about 4911s and 10s, respectively. Therefore, the optimization and real-time performance of DRL-based EMS can be guaranteed. • A VGG16-based neural network for road recognition is trained. • A complex and time-varying road model with idealized characteristics is established. • Aiming at achieving integrated control, a stereoscopic control network is modeled. • A DQN-based multi-objective control strategy is proposed. [ABSTRACT FROM AUTHOR]

This research proposes a reinforcement learning-based algorithm and a deep reinforcement learning-based algorithm for energy management of a series hybrid electric tracked vehicle. Firstly, the powertrain model of the series hybrid electric tracked vehicle (SHETV) is constructed, then the corresponding energy management formulation is established. Subsequently, a new variant of reinforcement learning (RL) method Dyna, namely Dyna-H, is developed by combining the heuristic planning step with the Dyna agent and is applied to energy management control for SHETV. Its rapidity and optimality are validated by comparing with DP and conventional Dyna method. Facing the problem of the "curse of dimensionality" in the reinforcement learning method, a novel deep reinforcement learning algorithm deep Q-learning (DQL) is designed for energy management control, which uses a new optimization method (AMSGrad) to update the weights of the neural network. Then the proposed deep reinforcement learning control system is trained and verified by the realistic driving condition with high-precision, and is compared with the benchmark method DP and the traditional DQL method. Results show that the proposed deep reinforcement learning method realizes faster training speed and lower fuel consumption than traditional DQL policy does, and its fuel economy quite approximates to global optimum. Furthermore, the adaptability of the proposed method is confirmed in another driving schedule. • The powertrain model of the series hybrid electric tracked vehicle is constructed. • A novel reinforcement learning-based energy management strategy is proposed. • The rapidity and optimality of the reinforcement learning method are validated. • A new optimization method is applied to update the weights of the neural network. • The proposed deep reinforcement learning method realizes better performance. [ABSTRACT FROM AUTHOR]