关键词: 光伏板, 缺陷检测, 卷积神经网络（CNN）, Transformer, 双分支

Abstract: Photovoltaic panels, as the core components of photovoltaic power generation systems, directly impact power generation efficiency and circuit safety. However, existing defect detection algorithms often fail to effectively integrate the advantages of convolutional neural network (CNN) and Transformers, limiting the overall model performance. To address this issue, this paper proposes the global and local feature enhanced YOLOX (GLF-YOLOX) algorithm. Firstly, the method employs a dual-branch backbone network that combines the local feature extraction strength of CNN with the global contextual modeling capability of Transformers, effectively capturing both fine-grained details and global information. Additionally, a global and local enhanced attention mechanism is introduced to dynamically fuse global and local features, improving the model’s focus on target regions and enhancing detail representation. Furthermore, a Transformer encoder layer-based detection head is designed to refine global feature modeling and optimize feature expression, significantly enhancing classification accuracy. Experimental results demonstrate that GLF-YOLOX achieves outstanding performance, with a mean average precision (mAP) improvement of approximately 4.5% over mainstream object detection methods, validating its effectiveness and superiority.

Key words: photovoltaic panel, defect detection, convolutional neural network (CNN), Transformer, double branch