Scholarlink

<p>Conveyor belts in coal mines are critical for coal extraction and safety. Detecting foreign objects in low-light underground environments is challenging. This paper presents an enhanced foreign body detection algorithm using an improved Dual-Model Low-Light Enhancement Algorithm (DLEA) and a lightweight Star Attention Region-based Convolutional Detection Transformer (SARC-DETR). The DLEA improves image quality in low-light conditions, while SARC-DETR, with its StarNet backbone and Efficient Additive Attention mechanism, reduces computational costs without compromising accuracy. A lightweight dynamic group efficient module network is proposed for optimized feature extraction, and the CIoU loss function further enhances positioning accuracy. Experimental results demonstrate a 4.7% precision improvement, a 2.7% increase in average precision, a 47.01% reduction in parameters, and an inference speed of 97.3 FPS. This approach significantly boosts detection accuracy and real-time performance in coal mine conveyor belt foreign object detection.</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

Enhanced foreign body detection on coal mine conveyor belts using improved DLEA and lightweight SARC-DETR model

  • 期刊论文

  • 发表日期：

    2025年3月4日

查看全文
  • Yan Hong,
  • Lei Wang,
  • Jingming Su,
  • Yun Li,
  • Biquan Zhu,
  • Hantao Wang

摘要

Conveyor belts in coal mines are critical for coal extraction and safety. Detecting foreign objects in low-light underground environments is challenging. This paper presents an enhanced foreign body detection algorithm using an improved Dual-Model Low-Light Enhancement Algorithm (DLEA) and a lightweight Star Attention Region-based Convolutional Detection Transformer (SARC-DETR). The DLEA improves image quality in low-light conditions, while SARC-DETR, with its StarNet backbone and Efficient Additive Attention mechanism, reduces computational costs without compromising accuracy. A lightweight dynamic group efficient module network is proposed for optimized feature extraction, and the CIoU loss function further enhances positioning accuracy. Experimental results demonstrate a 4.7% precision improvement, a 2.7% increase in average precision, a 47.01% reduction in parameters, and an inference speed of 97.3 FPS. This approach significantly boosts detection accuracy and real-time performance in coal mine conveyor belt foreign object detection.