基于EP-YOLO v8的瓶栽金针菇最优抓取位置定位方法

Optimal Grabbing Position Localization Method for Bottle-planted Enoki Based on EP-YOLO v8

摘要:

针对工厂化瓶栽金针菇自动切根过程中，夹持末端因结构设计导致行程固定，进而影响抓取效果甚至切根质量的问题，本文基于YOLO v8（You only look once）构建改进的Enoki-pick_region-YOLO v8(EP-YOLO v8)，实现瓶栽金针菇整体及最佳受力区域（关键抓取区域）的精准定位与轮廓提取，保障抓取参数的可靠性。该方法在网络优化基础上，基于最小欧几里得距离（Euclidean distance，ED）构建掩膜关系归属与判断模型，明确金针菇菇体与关键抓取区域掩膜间父子关系并合并优化。通过解析合并前后关键抓取区域的相对位置编码，确定抓取参数并进行坐标转换，为建立末端控制映射模型实现末端机械手运动行程的精确控制提供基础。实验结果表明，本文所提算法的金针菇菇体掩膜识别精确率达99.3%，关键抓取区域掩膜识别精确率达99.6%。同时，对比发现掩膜质量得到了提高，获取的参数抓取区域宽度与实际宽度之间的误差仅为0.7%，抓取参数基本满足抓取条件，能有效实现最优抓取位置的精准识别与定位。

Abstract:

Aiming at the problem that in the automatic root cutting process of factory bottle-planted enokis, the stroke of the clamping end was fixed due to the structural design, which affected the gripping effect and even the quality of root cutting, an improved enoki-pick_region-YOLO v8 based on the you only look once (YOLO v8) was constructed, realized accurate positioning and contour extraction of the whole bottle-planted enoki as well as the optimal stress area (the key picking region). The accurate localization and contour extraction of the whole bottle-planted enoki and the best stress region guaranteed the reliability of the grasping parameters. On the basis of network optimization, a mask attribution and judgment model based on the minimum Euclidean distance (ED) was constructed, the parentchild relationship between the enoki body mask and the key region mask was clarified, and they were merged for optimization. By analyzing the relative position encoding of the key region before and after the merger, the grasping parameters were determined and converted into coordinates, which provided the basis for establishing the end control mapping model to realize the precise control of the end manipulator’s motion stroke. The experimental results showed that the algorithm achieved a mask recognition rate of up to 99.3% for the enoki body and 99.6% for the key picking region. At the same time, it was found that the quality of the mask was improved, and the error between the width of the picking area and the actual width of the acquired parameters was only 0.7%, and the grasping parameters basically satisfied the conditions of grasping, which effectively realized the accurate identification and localization of the optimal grasping position.