Scholarlink

<p>To address the problems of low efficiency and insufficient identification accuracy in manual sorting during raw material sorting prior to potato processing, an online detection and rejection system integrating machine vision and automated sorting technology was developed. First, the hardware structure and control architecture of an automatic sorting system with an embedded platform as the core were established, enabling the coordinated operation of image acquisition, target detection, and actuator control. Second, with YOLOv12 as the baseline model, an improved lightweight object detection model, YOLOv12m-SSE, was proposed to enhance feature representation capability and the robustness of target recognition under complex conditions. Compared with the baseline YOLOv12m, YOLOv12m-SSE improved mAP@0.5 by roughly 1.6 percentage points, reaching 95.3% in the experimental evaluation. Meanwhile, the model size was compressed to approximately 7.5&#xa0;M parameters, and real-time inference performance of about 29 FPS was achieved on the embedded platform. To verify its engineering feasibility, the model was implemented in a self-developed automatic sorting system and tested in real working environments. Experimental results demonstrated that the developed automatic sorting system achieved a processing capacity of approximately 1200 tubers/h while maintaining high sorting accuracy under different conditions, indicating good stability and engineering applicability. The results obtained in this work contribute to the development of automated raw material sorting systems for potato processing.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automatic Detection and Rejection of Defective Potatoes and Impurities Using Machine Vision and a Sorting System

  • 期刊论文

  • 发表日期：

    2026年6月18日

查看全文
  • Danyang Lv,
  • Tao Zhang,
  • Minsheng Wu,
  • Ranbing Yang,
  • Jian Zhang,
  • Qimin Liang,
  • Shuo Han

摘要

To address the problems of low efficiency and insufficient identification accuracy in manual sorting during raw material sorting prior to potato processing, an online detection and rejection system integrating machine vision and automated sorting technology was developed. First, the hardware structure and control architecture of an automatic sorting system with an embedded platform as the core were established, enabling the coordinated operation of image acquisition, target detection, and actuator control. Second, with YOLOv12 as the baseline model, an improved lightweight object detection model, YOLOv12m-SSE, was proposed to enhance feature representation capability and the robustness of target recognition under complex conditions. Compared with the baseline YOLOv12m, YOLOv12m-SSE improved mAP@0.5 by roughly 1.6 percentage points, reaching 95.3% in the experimental evaluation. Meanwhile, the model size was compressed to approximately 7.5 M parameters, and real-time inference performance of about 29 FPS was achieved on the embedded platform. To verify its engineering feasibility, the model was implemented in a self-developed automatic sorting system and tested in real working environments. Experimental results demonstrated that the developed automatic sorting system achieved a processing capacity of approximately 1200 tubers/h while maintaining high sorting accuracy under different conditions, indicating good stability and engineering applicability. The results obtained in this work contribute to the development of automated raw material sorting systems for potato processing.