Scholarlink

<p>Cutting uniformity significantly influences the subsequent growth and development of potato seed tubers; however, this critical factor is frequently overlooked in automated seed-cutting processes. Leveraging a self-developed intelligent seed-cutting machine, this study employed a machine vision-based approach to extract the potato centroid, serving as a positional reference to enhance cutting uniformity. Traditional machine vision methods in centroid recognition are prone to environmental interference in edge extraction, while conventional semantic segmentation algorithms suffer from limited accuracy and high computational demands, failing to meet real-time performance requirements. To address these limitations, this study proposes an improved DFANet model for potato centroid recognition. The improved DFANet model, integrated with the traditional moment feature calculation method, ensures precise extraction of potato centroids. Experimental results demonstrate that the improved DFANet model achieves an accuracy of 98.78%, a recall of 98.65%, a mean intersection-over-union (mIoU) of 97.48%, and an average pixel accuracy of 98.65%, with a compact model size of 5.09Mb. These metrics demonstrated its efficacy in practical seed-cutting applications.</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

Potato Centroid Recognition Method Based on Lightweight Convolutional Network

  • 期刊论文

  • 发表日期：

    2026年5月16日

查看全文
  • Huiwang Zhang,
  • Hua Huang,
  • Kang Gao,
  • Cundong Zhang,
  • Shijun Wang,
  • Yun Yue

摘要

Cutting uniformity significantly influences the subsequent growth and development of potato seed tubers; however, this critical factor is frequently overlooked in automated seed-cutting processes. Leveraging a self-developed intelligent seed-cutting machine, this study employed a machine vision-based approach to extract the potato centroid, serving as a positional reference to enhance cutting uniformity. Traditional machine vision methods in centroid recognition are prone to environmental interference in edge extraction, while conventional semantic segmentation algorithms suffer from limited accuracy and high computational demands, failing to meet real-time performance requirements. To address these limitations, this study proposes an improved DFANet model for potato centroid recognition. The improved DFANet model, integrated with the traditional moment feature calculation method, ensures precise extraction of potato centroids. Experimental results demonstrate that the improved DFANet model achieves an accuracy of 98.78%, a recall of 98.65%, a mean intersection-over-union (mIoU) of 97.48%, and an average pixel accuracy of 98.65%, with a compact model size of 5.09Mb. These metrics demonstrated its efficacy in practical seed-cutting applications.