基于机器视觉的深纹核桃定向破壳方法

Machine Vision-based Directional Shell-breaking Method for Deep-grooved Walnuts

摘要:

针对目前核桃机械破壳过程施载方向及破壳间距随机导致破壳率与高露仁率不平衡的问题，本文以云南深纹核桃为研究对象，提出一种基于机器视觉与自动控制协同的定向破壳方法。开展核桃物理特性分析试验确定破壳最优方向，并结合赫兹接触理论设计内凹角20°、端面外径24 mm且表面带阵列式环形筋条的破壳挤压头。通过引入轻量化共享细节增强检测头（Lightweight shared detail-enhanced convolutional detection，LSDECD）改进 YOLO v8n-Pose姿态估计模型，提升微小特征感知能力。对深纹核桃缝合线端点进行特征训练，改进后模型目标框平均精度均值（mAP50-95）达83.4%，特征点平均精度均值达99.5%，且训练帧率达322.6 f/s，实际应用场景帧率达103.7 f/s。通过缝合线端点像素坐标计算其连线与水平坐标轴角度，该角度通过信号转换实现破壳夹臂角度补偿控制。开展核桃破壳接触方式与加载方向双因素试验，试验结果表明，在最佳作业参数下破壳率达93.81%，高露仁率达92.70%，与随机方向使用平型挤压头挤压破壳相比，分别提升14.50、11.68个百分点。研究结果为深纹核桃自动化连续输送定向破壳装备研发提供理论支撑与技术参考。

Abstract:

Aiming to address the imbalance between the shell-breaking rate and the high-quality kernel rate caused by random loading directions and cracking clearance during the mechanical walnut cracking process, a directional shell-breaking method synergizing machine vision and automatic control was proposed for Yunnan deep-grooved walnuts.Firstly, based on physical characteristic tests of the walnuts, the optimal shell-breaking direction was determined.Subsequently, guided by Hertzian contact theory, a customized concave shell-breaking indenter was designed, featuring a 20° concave angle, a 24 mm outer end-face diameter, and arrayed annular ribs on its surface.Secondly, to improve the perception of minute features, an improved YOLO v8n Pose pose estimation model was developed by introducing a lightweight shared detail-enhanced convolutional detection (LSDECD) head.Trained specifically on the suture endpoints of deep-grooved walnuts, a mean average precision ( mAP50 95 ) of 83.4% for bounding boxes and 99.5% for keypoints was achieved by the improved model.Furthermore, an ultra-high training frame rate of 322.6 f / s was demonstrated, with a real-time detection speed of 103.7 f / s in practical application scenarios.Next, the angle between the line connecting the suture endpoints and the horizontal axis was calculated by using their pixel coordinates, which was then converted into a control signal to achieve angle-compensation control of the shell-breaking clamping arms.Finally, a two-factor experiment on contact mode and loading direction was conducted.The experimental results showed that under the optimal operating parameters, a shell-breaking rate of 93.81% and a high-quality kernel rate of 92.70% were obtained.Compared with random-direction cracking with a flat indenter,these two indicators were improved by 14.50 percentage points and 11.68 percentage points, respectively.Thekernel-shell separation quality was effectively improved by this research, providing theoretical support and technical references for the development of automated, continuous-conveying, directional shell-breaking equipment for deep-grooved walnuts.