Scholarlink

<p>To address the low accuracy, slow speed, and large model size of traditional gangue detection systems, an improved MSRCR algorithm combined with adaptive gamma correction was first applied to enhance image quality. The YOLOv8n detection algorithm was further optimized through the integration of the FasterNet network. Specifically, the original C2F module in the Backbone layer was replaced with the more efficient C2F-Faster module, which reduced model size and accelerated inference speed. In addition, an efficient multi-scale attention module was incorporated into the C2F-Faster modules at the P3, P4, and P5 layers to enhance feature extraction and recognition. The Neck layer was further optimized by employing GSConv and VOV-GSCSP modules, thereby improving accuracy and reducing model complexity.Experimental results indicated that the optimized lightweight model required only 6.6 GFLOPs, representing 81.5% of the computational cost of the original model. The model achieved 98.7% accuracy, 97.8% recall, and 98.9% mean average precision, representing improvements of 1.4%, 3.5%, and 0.7%, respectively, compared with the baseline model. These results confirmed that the proposed method effectively balanced detection accuracy, speed, and computational efficiency, thereby making it suitable for real-time gangue detection. Moreover, the model also demonstrated robust recognition performance under varying lighting conditions and environments, exhibiting strong generalization and adaptability across diverse application scenarios.</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 YOLO architecture for real-time coal gangue detection with lightweight design

  • 期刊论文

  • 发表日期：

    2025年9月22日

查看全文
  • Wen Li,
  • Yan Hong,
  • Ruixian Pan,
  • Jingming Su,
  • Lei Wang,
  • Shiqiang Bao

摘要

To address the low accuracy, slow speed, and large model size of traditional gangue detection systems, an improved MSRCR algorithm combined with adaptive gamma correction was first applied to enhance image quality. The YOLOv8n detection algorithm was further optimized through the integration of the FasterNet network. Specifically, the original C2F module in the Backbone layer was replaced with the more efficient C2F-Faster module, which reduced model size and accelerated inference speed. In addition, an efficient multi-scale attention module was incorporated into the C2F-Faster modules at the P3, P4, and P5 layers to enhance feature extraction and recognition. The Neck layer was further optimized by employing GSConv and VOV-GSCSP modules, thereby improving accuracy and reducing model complexity.Experimental results indicated that the optimized lightweight model required only 6.6 GFLOPs, representing 81.5% of the computational cost of the original model. The model achieved 98.7% accuracy, 97.8% recall, and 98.9% mean average precision, representing improvements of 1.4%, 3.5%, and 0.7%, respectively, compared with the baseline model. These results confirmed that the proposed method effectively balanced detection accuracy, speed, and computational efficiency, thereby making it suitable for real-time gangue detection. Moreover, the model also demonstrated robust recognition performance under varying lighting conditions and environments, exhibiting strong generalization and adaptability across diverse application scenarios.