水培生菜低损采收柔性抓取参数视觉检测方法

Visual Detection Method of Flexible Grasping Parameters for Hydroponic Lettuce Low-damage Harvesting

摘要:

水培生菜是植物工厂主要栽培作物，菜叶柔嫩且分散交错，机械采收易出现菜叶损伤等问题。本文提出水培生菜叶展及叶面交错点视觉检测方法，并用于生菜柔性抓取参数调控。开展了基于水培生菜边缘轮廓提取的叶展视觉检测方法研究，实现相邻水培生菜重叠状态下叶展视觉检测。为提升水培生菜叶面交错点识别效果，分析了YOLO v5s、YOLO v5s-SimAM、YOLO v5s-SENet、YOLO v5s-CA对叶面交错点识别准确率等参数的影响，并结合生菜叶展检测结果筛选叶面交错点，获取生菜柔性抓取角度。结果表明：水培生菜叶展视觉检测相对误差平均值为2.46%；YOLO v5-CA模型对叶面交错点视觉识别效果最佳，识别准确率为94.1%,召回率为91.0%，平均精度均值为93.8%。在水培生菜特征视觉检测基础上，开展了采收验证试验并利用高速摄影进行分析，试验结果表明，采收成功率为97.23%，采收损伤程度为4.08%，实现了水培生菜低损伤柔性采收。

Abstract:

Hydroponic lettuce is the main crop in plant factories. However, mechanical harvesting currently causes serious leaf damage. The key problem of low-damage harvesting in plant factories was solved. A visual detection method for lettuce leaf expansion and overlapping points to control flexible grasping parameters was proposed, thereby improving harvesting quality. A vision-based leaf expansion detection method was carefully tested that used edge contour extraction to accurately measure leaf expansion, even when adjacent lettuce plants overlapped. The results were used to adjust flexible gripper diameters precisely. To optimize overlapping point recognition, comprehensive experiments were conducted by comparing YOLO v5s, YOLO v5s-SimAM, YOLO v5s-SENET, and YOLO v5s-CA models, with detailed analysis of their respective impacts on recognition accuracy parameters. Subsequently, based on the leaf expansion detection results, optimal overlapping points were screened. The positional information of these selected overlapping points was then used to calculate the most appropriate grasping angles for the robotic manipulator. Experimental results demonstrated that the visual detection system achieved an average relative error of merely 2.46% for leaf expansion measurement. Moreover, the YOLO v5s-CA model delivered superior performance in overlapping point recognition with 94.1% accuracy, 91.0% recall, and 93.8% mAP. Subsequent harvesting validation tests confirmed the effectiveness of this method, attaining a remarkable 97.23% success rate while maintaining minimal damage at just 4.08%, ultimately realizing high-quality, low-damage flexible harvesting of hydroponic lettuce.