Scholarlink

<p>The problems such as target occlusion and complex background in the defect images of photovoltaic modules make it difficult for the existing detection networks to accurately detect module defects. Therefore, this paper proposes a deep learning algorithm based on depthwise-separable convolution, differentiated feature representation, and self-focus optimization improvement, which is abbreviated as YOLOv11n-sps. Firstly, the backbone of YOLOv11n is optimized through the StarNet lightweight network, combined with the depth-separable convolution and channel shuffle mechanism, significantly reducing the number of parameters in the network and the computational load during the feature extraction process. Secondly, by integrating Partial Convolution (PConv) to improve the neck, the effective extraction ability for multi-scale features is maintained, and the number of model parameters and computational load is effectively reduced. Finally, the self-attention mechanism and the convolutional structure are integrated to optimize the head, thereby enhancing the aggregated semantic information at each scale within the receptive field range and effectively improving the detection effect of small-scale defects. The experimental results show that on the dataset of this paper, the proposed YOLOv11n-sps is compared with algorithms such as SSD, Faster R-CNN, and the YOLO series. The defect detection accuracy of mAP50 has been improved by at least 1.8%, and the number of parameters and computational load has been reduced by 1.23M and 1.6GFLOPs, respectively, compared to YOLOv11n, providing strong technical support for the intelligent operation and maintenance of photovoltaic power.</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

Research on defect detection of photovoltaic modules based on an improved deep learning framework

  • 期刊论文

  • 发表日期：

    2026年5月18日

查看全文
  • Chunsong Du,
  • Qian Zhang,
  • Buyang Qi,
  • Shaowei Liang,
  • Zhenghui Zhao

摘要

The problems such as target occlusion and complex background in the defect images of photovoltaic modules make it difficult for the existing detection networks to accurately detect module defects. Therefore, this paper proposes a deep learning algorithm based on depthwise-separable convolution, differentiated feature representation, and self-focus optimization improvement, which is abbreviated as YOLOv11n-sps. Firstly, the backbone of YOLOv11n is optimized through the StarNet lightweight network, combined with the depth-separable convolution and channel shuffle mechanism, significantly reducing the number of parameters in the network and the computational load during the feature extraction process. Secondly, by integrating Partial Convolution (PConv) to improve the neck, the effective extraction ability for multi-scale features is maintained, and the number of model parameters and computational load is effectively reduced. Finally, the self-attention mechanism and the convolutional structure are integrated to optimize the head, thereby enhancing the aggregated semantic information at each scale within the receptive field range and effectively improving the detection effect of small-scale defects. The experimental results show that on the dataset of this paper, the proposed YOLOv11n-sps is compared with algorithms such as SSD, Faster R-CNN, and the YOLO series. The defect detection accuracy of mAP50 has been improved by at least 1.8%, and the number of parameters and computational load has been reduced by 1.23M and 1.6GFLOPs, respectively, compared to YOLOv11n, providing strong technical support for the intelligent operation and maintenance of photovoltaic power.