Optimizing water and fertilizer management is crucial for improving cotton yield and quality. However, reliable and generalizable models for quickly and accurately estimating cotton canopy leaves water and nutritional status at a low cost throughout the entire growth stage are scarce. Therefore, this study aims to construct the generalization and adaptation retrieval model of cotton canopy leaf nitrogen content (LNC) and equivalent water thickness (EWT) based on PROSAIL, hyperspectral reconstruction with UAV multispectral imagery and module transfer learning. In the hyperspectral reconstruction module, the new hyperspectral reconstruction model (swinT-HSCNN) based on multispectral showing superior performance in reducing pixel-scale systematic errors and effectively captured spectral variations than HSCNN+ and MST++ model. In the PROSAIL module, the proposed Original-E2DCOS method demonstrated greater sensitivity to spectral response characteristics, especially for parameters and bands with low correlation values, and three bands (702 nm, 762 nm, and 938 nm) were selected as the sensitive bands corresponding to chlorophyll content (Cab) and equivalent water thickness (Cw) of cotton. The improved PROSAIL with hyperparameter optimization based on full spectrum and multispectral band shown better fitting performance than the model based on sensitive bands, and achieved high accuracy on simulated data, with R 2 values exceeding 0.98 for both Cab and Cw. Moreover, the new developed modular transfer learning retrieval model of cotton canopy water and nitrogen content through PROSAIL model and hyperspectral reconstruction with UAV multispectral imagery achieved good inversion accuracy with R 2 of 0.83, 0.85, RMSE of 0.0048, 0.0052, for LNC and EWT, respectively after verifying with actual experiment data. In summary, the proposed modular transfer learning retrieval model of cotton canopy water and nitrogen content integrates physical constraints into retrieval models, which enhancing their accuracy and generalization capability, and providing valuable technical support for precision agriculture in cotton production across different regions. • A modular transfer learning model integrates PROSAIL and hyperspectral reconstruction. • Achieves high accuracy for LNC and EWT estimation across diverse environments. • Combines UAV multispectral data with physical constraints for crop monitoring. • Reduces reliance on expensive hyperspectral sensors, ensuring cost-effectiveness. • Validated on multi-regional datasets, demonstrating scalability and generalizability.

Computer vision offers significant potential for the continuous, stress-free, and cost-effective monitoring of animal behavior, yet its application in goat farming remains limited. In this study, a Multi-Object Detection (MOD) model was developed to classify goat behavior into four categories: eating, standing, drinking, and lying. Zenithal videos were recorded under both light and dark conditions on an experimental goat farm, resulting in 10,740 labelled annotations used to train, validate, and test 13 models leveraging Transformer- and CNN-based pretrained architectures. YOLO-based models (YOLOv8 and YOLOX) achieved the highest overall performances across both large and lightweight versions, demonstrating high detection capability and potential suitability for hardware-constrained scenarios. YOLOX-based MOD model is preferred for goat behavior detection due to its superior classification accuracy, fast inference speed, and fully open-source license, enabling flexible customization, deployment, and reproducibility. Other models, particularly DAB-DETR and H-DINO, underperformed, especially in detecting drinking behavior, which represents the most challenging class due to its visual similarity with standing, class imbalance, and fisheye distortion effects that affect the frame regions where drinkers are located. Mitigation strategies, including focal loss and distortion correction, improved detection accuracy for this class and reduced performance variability. The developed MOD model can be deployed for continuous group-level monitoring of goats, paving the way for scalable and efficient solutions for advanced behavioral analyses. Future works will focus on integrating tracking algorithms for animal-level insights, as well as on evaluating model generalizability across different farming conditions and goat breeds. • Computer vision applications for goat farming are still very limited in literature. • Various models for goat behavior detection were developed and tested. • YOLOX achieved a mAP @ 0.50 of 0.957 with strong performances across the classes. • Drinking detection is hard due to visual similarity, class imbalance, distortion. • Frame undistortion improves drinking detection performance and reduces variability.

Maize seedling count and leaf age are critical indicators of early growth status, essential for effective field management and breeding variety selection. Traditional field monitoring methods are time-consuming, labor-intensive, and prone to subjective errors. Recently, deep learning-based object detection models have gained attention in crop seedling counting. However, many of these models exhibit high computational complexity and implementation costs, making field deployment challenging. Moreover, maize leaf age monitoring in field environments is barely investigated. Therefore, this study proposes two lightweight models, YOLOv8n-Light-Pruned (YOLOv8n-LP) and YOLOv11n-Light-Pruned (YOLOv11n-LP), for monitoring maize seedling count and leaf age in field RGB images. Our proposed models are improved from YOLOv8n and YOLOv11n by incorporating the DAttention mechanism, an improved BiFPN, an EfficientHead, and layer-adaptive magnitude-based pruning. The improvement in model complexity and model efficiency was significant, with the number of parameters reduced by over 73 % and model efficiency upgraded by up to 42.9 % depending on the device computation power. High accuracy was achieved in seedling counting (YOLOv8n-LP/ YOLOv11n-LP: AP = 0.968/0.969, R 2 = 0.91/0.94, rRMSE = 6.73 %/5.59 %), with significantly reduced model size (YOLOv8n-LP/ YOLOv11n-LP: parameters = 0.8 M/0.7 M, trained model size = 1.8 MB/1.7 MB). The robustness was validated across datasets with varying leaf ages (rRMSE = 4.07 % – 7.27 %), resolutions (rRMSE = 3.06 % – 6.28 %), seedling compositions (rRMSE = 1.09 % – 9.29 %), and planting densities (rRMSE = 3.38 % – 10.82 %). Finally, by integrating plant counting and leaf age estimation, the proposed models demonstrated high accuracy in leaf age detection using near-ground images (YOLOv8n-LP/ YOLOv11n-LP: rRMSE = 5.73 %/7.54 %) and UAV images (rRMSE = 9.24 %/14.44 %). The results demonstrate that the proposed models excel in detection accuracy, deployment efficiency, and adaptability to complex field environments, providing robust support for practical applications in precision agriculture. • Compact model size (1.7 MB), with only 0.7 million parameters. • One model counts both maize seedlings and leaves in the complex field environment. • Robust across data collection platforms, growth stages, image resolutions, seedling compositions, and generalizable to a lower planting density. • Model efficiency improved by 40 % on devices with low computation power.

Soil properties are vital for soil health, fertility, and crop productivity, yet traditional methods like the Partial Least Squares Regression (PLSR) predict individual properties without capturing their interactions or impacts on crop growth. This study presents the Soil Knowledge-Guided Multi-task Transformer (KGMT) model, which integrates deep learning and multi-task learning to simultaneously predict multiple soil properties and crop traits using UAV hyperspectral imagery, eliminating complex manual feature extraction. By embedding soil knowledge, KGMT improves prediction accuracy and generalization over PLSR and baseline multi-task models. Data from two agricultural plots in Bayannur City, Inner Mongolia, China, demonstrated KGMT's superior performance in predicting six soil properties (alkalinity, salinity, organic matter, potassium, sodium, calcium) and two crop traits (leaf area index, aboveground biomass), achieving up to 20% lower normalized root mean square error. For instance, KGMT yielded higher R 2 values for alkalinity (0.57) and organic matter (0.57) compared to PLSR (0.46 and 0.40, respectively). The model also enhanced cross-region generalization. Additionally, a soil early warning framework was developed, linking pre-planting soil conditions with crop traits to detect potential growth limitations early, enabling proactive management. This approach integrates spectral data with domain knowledge, offering a robust solution for soil and crop management. It contributes to precision agriculture by optimizing soil conditions and promoting sustainable land use practices, highlighting the value of advanced deep learning in agricultural applications. • Innovative Model for Multi-task Prediction : The Soil Knowledge-Guided Multi-task Transformer (KGMT) combines deep learning and domain knowledge to simultaneously predict multiple soil properties and crop growth traits directly from UAV hyperspectral imagery, eliminating the need for manual feature extraction. • Enhanced Prediction Accuracy and Generalization : KGMT outperformed traditional PLSR and baseline multi-task models in predicting six soil properties (e.g., alkalinity, salinity, organic matter, potassium, sodium, and calcium) and crop traits such as leaf area index (R 2 = 0.75) and aboveground biomass (R 2 = 0.82), achieving up to a 20% reduction in normalized root mean square error (nRMSE). • Proactive Soil and Crop Management : A novel soil early warning approach integrates soil properties from the bare soil stage with crop traits during growth, enabling early detection of potential limitations to crop growth and supporting precision agriculture for sustainable land use.

This study optimizes UAV sampling methods and path planning, and constructs the AIC-UNet (Improved ASPP-InceptionV4-CBAM UNet) model to mitigate the interference of illuminance variations with residual plastic film (RPF) extraction from cotton field images. This achieves rapid and automatic monitoring of RPF pollution in pre-sowing cotton fields. Current research faces challenges such as low RPF extraction accuracy, poor alignment between the UAV sampling range and the plot boundaries, and low path planning efficiency. To tackle these challenges, this study optimizes the UAV sampling process and innovatively proposes an RPF extraction model. Firstly, this study designs a UAV adaptive sampling method by combining the geometric center-based radiation method and the Held-Karp algorithm. This achieves dynamic adjustment of sampling intervals and accurate matching with plot boundaries, providing high-quality image samples for AIC-UNet modeling. Then, this study constructs the AIC-UNet model that integrates the improved ASPP, CBAM, and InceptionV4 structure to efficiently extract RPF features under complex illuminance conditions. The results show that the proposed scheme not only enhances RPF pollution assessment accuracy under different illuminance conditions but also significantly enhances the monitoring efficiency and practicality. The AIC-UNet achieves a mIoU of 88.59% for RPF extraction under different illuminance conditions, and the R 2 reaches 0.968 in the linear fitting with manually labelled values. Additionally, the average UAV sampling path planning time is only 0.06963 s, and the average D (superiority degree of the path) reaches 100%, significantly enhancing the efficiency of RPF monitoring. This study provides technical support for farmland RPF pollution assessment and contributes to sustainable agricultural development. • A multi-illumination environmental dataset was created via UAV low-altitude imaging. • Geometric-center radiation sampling was used for adaptive UAV sampling tuning. • Held-Karp was used for rapid planning of the shortest UAV sampling path. • AIC-UNet was proposed for residual film recognition and image segmentation. • Technical support was provided for sustainable agriculture development.

As precision agriculture evolves, unmanned ground vehicles (UGVs) have become an essential tool for improving weed management techniques, offering automated and targeted methods that obviously reduce the reliance on manual labor and blanket herbicide applications. Several papers on UGV-based weed control methods have been published in recent years, yet there is no explicit attempt to systematically study these papers to discuss these weed control methods, UGVs adopted, and their key components, and how they impact the environment and economy. Therefore, the objective of this study was to present a systematic review that involves the efficiency and types of weed control methods deployed in UGVs, including mechanical weeding, targeted herbicide application, thermal/flaming weeding, and laser weeding in the last 2 decades. For this purpose, a thorough literature review was conducted, analyzing 68 relevant articles on weed control methods for UGVs. The study found that the research focus on using UGVs in mechanical weeding has been more dominant, followed by target or precision spraying/ chemical weeding, with hybrid weeding systems quickly emerging. The effectiveness of UGVs for weed control is hinged on the accuracy of their navigation and weed detection technologies, which are influenced heavily by environmental conditions, including lighting, weather, uneven terrain, and weed and crop density. Also, there is a shift from using traditional machine learning (ML) algorithms to deep learning neural networks, including convolutional neural networks (CNNs) and recurrent neural networks (RNNs), for weed detection algorithm development due to their potential to work in complex environments. Finally, trials of most UGVs have limited documentation or lack extensive trials under various conditions, such as varying soil types, crop fields, topography, field geometry, and annual weather conditions. This review paper serves as an in-depth update on UGVs in weed management for farmers, researchers, robotic technology industry players, and AI enthusiasts, helping to further foster collaborative efforts to develop new ideas and advance this revolutionary technique in modern agriculture.

Verticillium wilt poses a severe threat to cotton growth and significantly impacts cotton yield. It is of significant importance to detect Verticillium wilt stress in time. In this study, the effects of Verticillium wilt stress on the microstructure and physiological indicators (SOD, POD, CAT, MDA, Chl a , Chl b , Chl ab , Car) of cotton leaves were investigated, and the feasibility of utilizing hyperspectral imaging to estimate physiological indicators of cotton leaves was explored. The results showed that Verticillium wilt stress-induced alterations in cotton leaf cell morphology, leading to the disruption and decomposition of chloroplasts and mitochondria. In addition, compared to healthy leaves, infected leaves exhibited significantly higher activities of SOD and POD, along with increased MDA amounts, while chlorophyll a nd carotenoid levels were notably reduced. Furthermore, rapid detection models for cotton physiological indicators were constructed, with the R p of the optimal models ranging from 0.809 to 0.975. Based on these models, visual distribution maps of the physiological signatures across cotton leaves were created. These results indicated that the physiological phenotype of cotton leaves could be effectively detected by hyperspectral imaging, which could provide a solid theoretical basis for the rapid detection of Verticillium wilt stress. • TEM shows cell damage and more starch granules in infected cotton leaves. • Infected leaves show higher SOD, POD, MDA, and lower Chl a , Chl b , Chl ab , Car. • Hyperspectral models detect and visualize antioxidant and pigment distributions. • SOD, POD, MDA levels are higher near veins, suggesting early disease response.

In the complex orchard environment, precise picking point localization is crucial for the automation of fruit picking robots. However, existing methods are prone to positioning errors when dealing with complex scenarios such as short peduncles, partial occlusion, or complete misidentification, which can affect the actual work efficiency of the fruit picking robot. This study proposes an enhanced picking point localization method based on semantic reasoning for complex picking scenarios in vineyard. It innovatively designs three modules: the semantic reasoning module (SRM), the ROI threshold adjustment strategy (RTAS), and the picking point location optimization module (PPOM). The SRM is applied to handle the scenarios of grape peduncles being obstructed by obstacles, partial misidentification of peduncles, and complete misidentification of peduncles. The RTAS addresses the issue of low and short peduncles during the picking process. Finally, the PPOM optimizes the final position of the picking point, allowing the robotic arm to perform the picking operation with greater flexibility. Experimental results show that SegFormer achieves an mIoU (mean Intersection over Union) of 84.54 %, with B_IoU and P_IoU reaching 73.90 % and 75.63 %, respectively. Additionally, the success rate of the improved fruit picking point localization algorithm reached 94.96 %, surpassing the baseline algorithm by 8.12 %. The algorithm's average processing time is 0.5428 ± 0.0063 s, meeting the practical requirements for real-time picking. • Propose a high-precision picking point localization method, meeting real-time needs. • Semantic reasoning achieves advanced inference from existing semantic information. • Three algorithms—SRM, RTAS, PPOM—improve picking point accuracy in complex scenarios.