Scholarlink

<p>In industrial steel bar detection, common challenges include image quality degradation due to complex lighting, missed detection of circular edge features, and dense occlusions. To address these issues, this paper proposes a dual-path optimized YOLO11n detection framework. Firstly, we design the Adaptive Residual Enhancer (ARE) module, which significantly enhances the detectability of features under low-light conditions through dynamic brightness adjustment and cascaded multi-residual blocks. Secondly, we introduce the Polar Coordinate Circular Dynamic Snaking Convolution (PCCDSConv), transforming the traditional rigid sampling convolution in Cartesian space into dynamic radius adjustment and chain-based snaking paths in polar coordinates, achieving geometric adaptability when capturing circular edge features. Experimental results show that the synergy of ARE and PCCDSConv significantly improves the mAP50-95 on the RebarDSC dataset, achieving a 2.3% improvement over YOLO11n and outperforming YOLOv10n, YOLOv5 and YOLO11n-PENet by 0.4%, 1.1%, and 2.7%, respectively. The proposed method demonstrates outstanding robustness and precision in environments with large lighting variations, steel bar stacking, and occlusions.</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

AP-YOLO11n: Rebar counting with adaptive residual and polar coordinate snaking convolution

  • 期刊论文

  • 发表日期：

    2025年10月18日

查看全文
  • Yadi Zhao,
  • Zhigang Lian,
  • Qi Cao

摘要

In industrial steel bar detection, common challenges include image quality degradation due to complex lighting, missed detection of circular edge features, and dense occlusions. To address these issues, this paper proposes a dual-path optimized YOLO11n detection framework. Firstly, we design the Adaptive Residual Enhancer (ARE) module, which significantly enhances the detectability of features under low-light conditions through dynamic brightness adjustment and cascaded multi-residual blocks. Secondly, we introduce the Polar Coordinate Circular Dynamic Snaking Convolution (PCCDSConv), transforming the traditional rigid sampling convolution in Cartesian space into dynamic radius adjustment and chain-based snaking paths in polar coordinates, achieving geometric adaptability when capturing circular edge features. Experimental results show that the synergy of ARE and PCCDSConv significantly improves the mAP50-95 on the RebarDSC dataset, achieving a 2.3% improvement over YOLO11n and outperforming YOLOv10n, YOLOv5 and YOLO11n-PENet by 0.4%, 1.1%, and 2.7%, respectively. The proposed method demonstrates outstanding robustness and precision in environments with large lighting variations, steel bar stacking, and occlusions.