Scholarlink

<p>During post-harvest handling, transportation, and sorting, fruits often suffer from hidden bruises that are difficult to detect immediately. This study investigated the early hidden bruise sensitivity of six common fruits in the near-infrared range (700–1100&#xa0;nm) and developed a portable detection device integrated with a near-infrared camera. A lightweight YOLO-Faster object detection algorithm, based on a YOLOv8 model, was proposed for efficient and accurate bruise detection. The network is optimized with lightweight Partial Convolution modules and Bi-level Routing attention mechanisms to reduce parameters and improve accuracy. Additionally, an adaptive image enhancement combination algorithm was proposed to enhance bruise regions and address environmental interference challenges. Results showed that the model achieves a mAP@50 of 97.4%, with parameters and computation reduced to 0.79&#xa0;M and 5.0 G, enabling real-time detection on embedded devices. This system addresses the current limitations related to cost, portability, and real-time performance, providing an innovative solution for on-site detection and advancing intelligent detection of hidden bruises on fruits.</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

Fruit early bruise detection using a portable device developed based on near-infrared imaging combined with proposed YOLO-Faster model

  • 期刊论文

  • 发表日期：

    2025年11月20日

查看全文
  • Dongjie Huo,
  • Xiaping Fu,
  • Hanxue Zhao,
  • Huizhang Wang

摘要

During post-harvest handling, transportation, and sorting, fruits often suffer from hidden bruises that are difficult to detect immediately. This study investigated the early hidden bruise sensitivity of six common fruits in the near-infrared range (700–1100 nm) and developed a portable detection device integrated with a near-infrared camera. A lightweight YOLO-Faster object detection algorithm, based on a YOLOv8 model, was proposed for efficient and accurate bruise detection. The network is optimized with lightweight Partial Convolution modules and Bi-level Routing attention mechanisms to reduce parameters and improve accuracy. Additionally, an adaptive image enhancement combination algorithm was proposed to enhance bruise regions and address environmental interference challenges. Results showed that the model achieves a mAP@50 of 97.4%, with parameters and computation reduced to 0.79 M and 5.0 G, enabling real-time detection on embedded devices. This system addresses the current limitations related to cost, portability, and real-time performance, providing an innovative solution for on-site detection and advancing intelligent detection of hidden bruises on fruits.