"Research on Tea Yield Detection Using YOLOv10 Based on UAV Remote Sensing"

"To address the limited accuracy and robustness of UAV-based tea-yield detection under complex plantation conditions where scale variation, severe background clutter, image blur, and canopy occlusion jointly degrade performance this study proposes an enhanced object detection network, termed HR-YOLOv10-S. Built upon YOLOv10, the proposed network incorporates shape weights and scale adjustment factors into the loss function to improve bounding-box regression matching and accelerate convergence. In addition, a Rectangular Self-Calibrated Module is introduced to strengthen spatial representation and boundary localization, while a Histogram Transformer is employed to suppress background interference and refine the characterization of canopy structures and leaf-level details. Experimental results show that HR-YOLOv10-S achieves Precision, Recall, mAP, and F1 scores of 91.11%,88.91%, 93.33%, and 89.99%, respectively, corresponding to improvements of 8.68%, 4.03%, 5.18%, and 6.36% over YOLOv10. Moreover, compared with representative detectors (YOLOv10, SSD, CornerNet, and RT-DETR), HR-YOLOv10-S improves Precision by 8.68%, 11.49%, 13.15%, and 9.19%; Recall by 4.03%, 11.83%, 13.55%, and 5.27%; F1 by 6.36%, 11.67%, 13.36%, and 7.22%; and mAP by 5.18%, 11.79%, 14.47%, and 6.65%, demonstrating consistently superior overall performance. These results indicate that the proposed method effectively enhances the accuracy and robustness of UAV remote-sensing\u2013based yield detection in complex tea-garden environments, providing an efficient and reliable solution for intelligent tea-yield monitoring."