Scholarlink

<p>To address the safety detection issue of workers using aerial cable transport devices in mines, this paper proposes a lightweight real-time detection algorithm named MOOD. Considering the challenge of balancing detection speed and accuracy in dynamic working conditions, this study redesigns the network model based on FasterNet, GSCSP, and ELAN-P, incorporating parts of the YOLOv7 Head structure. This redesign significantly reduces the number of parameters and computational cost, thereby improving detection speed. Furthermore, to compensate for the loss of feature extraction information caused by network lightweighting, the CA attention mechanism is introduced into the FasterNet module to enhance feature extraction capabilities and mitigate underground lighting interference. In addition, an adaptively designed spatial ASFF structure is reconstructed to further strengthen multi-level feature interactions, improving detection accuracy in complex backgrounds and multi-scale pedestrian. Experiments show that MOOD achieves a mean average precision (mAP) of 96.6% in mining scenarios, with an inference speed of 121 FPS and a model size of only 23.3&#xa0;MB. Compared to existing algorithms, MOOD demonstrates significant improvements in both detection accuracy and real-time performance, providing an effective technical solution for safety monitoring.</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

A lightweight network and adaptive spatial fusion-based detection algorithm for mine aerial passenger devices

  • 期刊论文

  • 发表日期：

    2025年5月28日

查看全文
  • Ruxin Gao,
  • Haiquan Jin,
  • Xinyu Li,
  • Tengfei Wang,
  • Qunpo Liu

摘要

To address the safety detection issue of workers using aerial cable transport devices in mines, this paper proposes a lightweight real-time detection algorithm named MOOD. Considering the challenge of balancing detection speed and accuracy in dynamic working conditions, this study redesigns the network model based on FasterNet, GSCSP, and ELAN-P, incorporating parts of the YOLOv7 Head structure. This redesign significantly reduces the number of parameters and computational cost, thereby improving detection speed. Furthermore, to compensate for the loss of feature extraction information caused by network lightweighting, the CA attention mechanism is introduced into the FasterNet module to enhance feature extraction capabilities and mitigate underground lighting interference. In addition, an adaptively designed spatial ASFF structure is reconstructed to further strengthen multi-level feature interactions, improving detection accuracy in complex backgrounds and multi-scale pedestrian. Experiments show that MOOD achieves a mean average precision (mAP) of 96.6% in mining scenarios, with an inference speed of 121 FPS and a model size of only 23.3 MB. Compared to existing algorithms, MOOD demonstrates significant improvements in both detection accuracy and real-time performance, providing an effective technical solution for safety monitoring.