Weed infestation contributes significantly to global agricultural yield loss and increases the reliance on herbicides, raising both economic and environmental concerns. Effective weed detection in agriculture requires high accuracy and architectural efficiency. This is particularly important under challenging field conditions, including densely clustered targets, small weed instances, and low visual contrast between vegetation and soil. In this study, we propose GTDR-YOLOv12, an improved object detection framework based on YOLOv12, tailored for real-time weed identification in complex agricultural environments. The model is evaluated on the publicly available Weeds Detection dataset, which contains a wide range of weed species and challenging visual scenarios. To achieve better accuracy and efficiency, GTDR-YOLOv12 introduces several targeted structural enhancements. The backbone incorporates GDR-Conv, which integrates Ghost convolution and Dynamic ReLU (DyReLU) to improve early-stage feature representation while reducing redundancy. The GTDR-C3 module combines GDR-Conv with Task-Dependent Attention Mechanisms (TDAMs), allowing the network to adaptively refine spatial features critical for accurate weed identification and localization. In addition, the Lookahead optimizer is employed during training to improve convergence efficiency and reduce computational overhead, thereby contributing to the model’s lightweight design. GTDR-YOLOv12 outperforms several representative detectors, including YOLOv7, YOLOv9, YOLOv10, YOLOv11, YOLOv12, ATSS, RTMDet and Double-Head. Compared with YOLOv12, GTDR-YOLOv12 achieves notable improvements across multiple evaluation metrics. Precision increases from 85.0% to 88.0%, recall from 79.7% to 83.9%, and F1-score from 82.3% to 85.9%. In terms of detection accuracy, mAP:0.5 improves from 87.0% to 90.0%, while mAP:0.5:0.95 rises from 58.0% to 63.8%. Furthermore, the model reduces computational complexity. GFLOPs drop from 5.8 to 4.8, and the number of parameters is reduced from 2.51 M to 2.23 M. These reductions reflect a more efficient network design that not only lowers model complexity but also enhances detection performance. With a throughput of 58 FPS on the NVIDIA Jetson AGX Xavier, GTDR-YOLOv12 proves both resource-efficient and deployable for practical, real-time weeding tasks in agricultural settings.

Increasing demand for sustainable agriculture necessitates precise and efficient crop management to minimize resource wastage and environmental impact. To improve the precision of pesticide application in tomato leaves, a real-time tomato leaf detection method using an improved YOLOv8 algorithm is proposed. The framework was developed by integrating Depthwise Grouped Convolutions and an AdamW optimizer to achieve both computational efficiency and precise detection capabilities. The integration of SE_Block further enhanced feature representation by adaptively recalibrating channel-wise attention, improving detection accuracy and robustness. The algorithm was labeled and trained by using a diverse dataset of 1500 tomato leaf images consisting of four labels (All, Green Tomato, Downy Mildew, and Powdery Mildew), capturing variations in disease types, lighting conditions, and leaf orientations, enabling robust detection performance across real-world scenarios. The incorporation of Depthwise Grouped Convolutions into YOLOv8 reduced the computational complexity, enabling faster inference speed without sacrificing detection accuracy. Additionally, the AdamW optimizer enhanced the model convergence during training, ensuring robustness and stability. Compared with the original algorithm, the improved YOLOv8 achieved a significant performance improvement, with model precision (P%) increasing from 83.5% to 85.7% (2.2% increase), recall (R%) improving from 70.4% to 72.8% (2.4% increase), and mAP@0.5 improving from 75.7% to 79.8% (4.1% increase). mAP@0.5:0.95 also saw an improvement, rising from 44.2% to 51.6% (7.4% increase). Furthermore, the F1 score increased from 76.4% to 78.6% (2.2% increase), demonstrating enhanced overall detection accuracy. The system was deployed on the Spraying Robot LPE-260 to enable real-time, automated pesticide application in controlled environments. The improved detection framework ensures the targeted spraying of diseased tomato leaves, significantly reducing chemical usage and minimizing overspray. This system ensures that pesticide is sprayed exclusively on the diseased areas of tomato leaves, further minimizing chemical usage and overspray. It demonstrates the potential of computationally efficient deep learning techniques to address key challenges in precision agriculture, advancing scalable, sustainable, and resource-efficient crop management solutions.

The Rapidly-exploring Random Tree Star (RRT*) algorithm is widely applied in robotic manipulator path planning, yet it does not directly consider motion control, where abrupt changes may cause shocks and vibrations, reducing accuracy and stability. To overcome this limitation, this paper proposes the Kinematically Smoothed, dynamically Biased Bidirectional Potential-guided RRT* (KSBB-P-RRT*) algorithm, which unifies path planning and motion control and introduces three main innovations. First, a fast path search strategy on the basis of Bi-RRT* integrates adaptive sampling and steering to accelerate exploration and improve efficiency. Second, a triangle-inequality-based optimization reduces redundant waypoints and lowers path cost. Third, a kinematically constrained smoothing strategy adapts a Jerk-Continuous S-Curve scheme to generate smooth and executable trajectories, thereby integrating path planning with motion control. Simulations in four environments show that KSBB-P-RRT* achieves at least 30% reduction in planning time and at least 3% reduction in path cost, while also requiring fewer iterations compared with Bi-RRT*, confirming its effectiveness and suitability for complex and precision-demanding applications such as agricultural robotics.

Precision spraying in vineyards demands accurate and efficient plant disease detection to minimize chemical waste and environmental contamination. This paper introduces DEPA-VINE, an improved lightweight YOLOv10-based detection framework designed for real-time grapevine leaf disease detection and autonomous spraying control. The network replaces the standard YOLOv10 backbone layers with stacked Mobile Inverted Bottleneck Convolution (MBConv) blocks and introduces a Dynamic Convolutional Aggregation (DCA) module at the input stage to enhance low-level feature extraction under motion blur and occlusion. The backbone concludes with a Spatial Pyramid Pooling and Multi-scale Enhanced Fusion (SPPMEF) block to strengthen global context modeling, while the head integrates a C2f-Extended LSK-enhanced Transformer (C2f-XLT) for improved semantic alignment and cross-scale feature fusion. The model is further optimized using the AdaBelief optimizer, which stabilizes convergence and improves generalization in lightweight settings. The proposed framework achieves 83.4% precision, a 66.8% F1-score, and 70.2% mAP@0.5 with only 6.1 GFLOPs, running at an average inference speed of 63.7 FPS on the Rockchip RK3588 embedded platform. Comparative evaluations against YOLOv5n, YOLOv7t, YOLOv6n, YOLOv12n, RetinaNet, and ATSS confirm superior accuracy-efficiency trade-offs and robustness in challenging vineyard conditions. Field experiments using an autonomous robotic sprayer demonstrate a 27% increase in droplet coverage on diseased foliage and a 90% reduction in unintended spraying on healthy leaves, leading to substantial pesticide savings. These results confirm the practicality of the proposed framework for real-time selective pesticide application, enabling more sustainable and automated vineyard management. The implementation code and dataset will be publicly available.