Scholarlink

<p>Aiming to address the challenges of feature extraction in low-light conditions, complex models, and high-frequency vibrations within the domain of mine wire rope vibration monitoring, this paper proposes a lightweight network model, LCE-YOLOv10, with light compensation enhancement capabilities. Firstly, the Self-Calibrating Illumination Network (SCINet) module was employed to enhance the capability of low-light image improvement. Secondly, an efficient lightweight module named Cross Stage Partial Heterogeneous Convolution (CSPHet) was developed by integrating the “split-merge” strategy of the Cross Stage Partial (CSP) structure with Heterogeneous Convolution (HetConv). Finally, the Polarized Self-Attention (PSA) mechanism module was replaced with the Spatial and Channel Synergistic Attention (SCSA) mechanism module, thereby strengthening the model’s robustness against illumination variations and high-frequency vibrations through dual-dimensional feature interaction. The verification conducted on our self-constructed dataset of 3859 samples shows that the improved LCE-YOLOv10 model achieves a mAP of 99.3%, representing a 0.7% increase compared to the 98.6% mAP of the YOLOv10n model. Meanwhile, the number of parameters decreased from 2.71&#xa0;M to 2.15&#xa0;M, a reduction of 20.66%, the FPS increased from 71 to 107, an increase of 50.70%. The outcome conclusively shows that the model holds significant advantages for real-time monitoring tasks involving wire rope vibration detection in machine vision measurements at edge devices, thereby providing robust support for relevant technical 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

Detection of wire rope vibration targets in low-light mine environments based on LCE-YOLOv10

  • 期刊论文

  • 发表日期：

    2025年9月9日

查看全文
  • Xin Hu,
  • Ziqiang Xie,
  • Chenglong Yang,
  • Chao Duo,
  • Wenshan Wang

摘要

Aiming to address the challenges of feature extraction in low-light conditions, complex models, and high-frequency vibrations within the domain of mine wire rope vibration monitoring, this paper proposes a lightweight network model, LCE-YOLOv10, with light compensation enhancement capabilities. Firstly, the Self-Calibrating Illumination Network (SCINet) module was employed to enhance the capability of low-light image improvement. Secondly, an efficient lightweight module named Cross Stage Partial Heterogeneous Convolution (CSPHet) was developed by integrating the “split-merge” strategy of the Cross Stage Partial (CSP) structure with Heterogeneous Convolution (HetConv). Finally, the Polarized Self-Attention (PSA) mechanism module was replaced with the Spatial and Channel Synergistic Attention (SCSA) mechanism module, thereby strengthening the model’s robustness against illumination variations and high-frequency vibrations through dual-dimensional feature interaction. The verification conducted on our self-constructed dataset of 3859 samples shows that the improved LCE-YOLOv10 model achieves a mAP of 99.3%, representing a 0.7% increase compared to the 98.6% mAP of the YOLOv10n model. Meanwhile, the number of parameters decreased from 2.71 M to 2.15 M, a reduction of 20.66%, the FPS increased from 71 to 107, an increase of 50.70%. The outcome conclusively shows that the model holds significant advantages for real-time monitoring tasks involving wire rope vibration detection in machine vision measurements at edge devices, thereby providing robust support for relevant technical applications.