Authors

Abstract

Current research on shrub stubble cutting machines primarily focuses on mechanical structure design and operational performance, whereas shrub localization and cutting status still rely on manual judgment. To address this issue, an intelligent desert shrub stubble cutting machine integrating structural design with vision‑assisted operation was developed. The overall structure and key component parameters were determined through theoretical analysis, and the disc cutting process was validated via Workbench simulation. The YOLOv5s model incorporating a Focal‑CIoU loss function was adopted to enhance shrub detection under complex background conditions. Image processing and binocular vision were employed to obtain the three‑dimensional coordinates of shrub roots. Experimental results demonstrate that the cutting device meets the operational requirements; the improved model achieves an accuracy of 91.9%, a recall of 96.5%, and an mAP of 98.3%; and the maximum relative errors for root depth and height measurements are 3.19% and 6.51%, respectively, satisfying the detection and localization requirements for shrub stubble cutting.

Abstract in Chinese

当前灌木平茬机的研究主要集中于机械结构设计与作业性能，而灌木定位和切割状态仍依赖人工判断。针对该问题，本文研制了一种将结构设计与视觉辅助作业相结合的智能荒漠灌木割茬机。通过理论分析确定了整机结构和关键部件参数，并利用Workbench仿真验证了圆盘切割过程。采用引入Focal‑CIoU损失函数的YOLOv5s模型，以增强复杂背景条件下的灌木检测能力。利用图像处理和双目视觉获取了灌木根部的三维坐标。实验结果表明：切割装置满足作业要求；改进模型的准确率达到91.9%，召回率为96.5%，mAP为98.3%；根部深度和高度测量的最大相对误差分别为3.19%和6.51%，满足灌木平茬检测与定位要求。