基于改进YOLOv13模型的光伏板缺陷检测方法研究

Research on a Photovoltaic Panel Defect Detection Method Based on an Improved YOLOv13

摘要: 针对目前光伏板缺陷检测中仍存在小目标检测率低、检测速度慢以及适应性差等问题，提出了一种基于YOLOv13n的改进光伏板缺陷检测模型YOLOv13n-PV。在YOLOv13的Backbone网络结构中引入了CA注意力机制，这种机制使得算法能够更加专注于目标的位置信息和类别判定，从而有效地提升了对于小目标和密集目标的特征提取能力。同时，对于损失函数进行优化，采用类别加权和小目标加权的双重加权机制，增强模型对缺陷的敏感度和加强对小缺陷的检测效果。通过各项试验结果表明，本文提出的算法模型在测试数据集下的平均准确率、召回率分别为94.1%和93.6%。分别优于原始YOLOv13n模型算法的93.2%和91.7%。且模型的计算量没有过多增加，在提升检测性能的同时能够兼顾算法计算效率，因此可以快速地、准确地实现光伏板的缺陷检测，为新能源系统中的光伏法发电提供技术支持。

Abstract: To address the persistent issues in photovoltaic panel defect detection—namely the low detection rate for small targets, slow inference speed, and poor adaptability—we propose an improved defect detection model, YOLOv13n-PV, based on YOLOv13n. In the YOLOv13 backbone, we incorporate the Coordinate Attention (CA) mechanism, which enables the algorithm to focus more effectively on spatial localization and category discrimination, thereby enhancing feature extraction for small and densely distributed targets. Meanwhile, the loss function is optimized by adopting a dual-weighting mechanism that combines class weighting and small-object weighting. This approach enhances the model’s sensitivity to defects and strengthens the detection performance for small defects. The results of various experiments show that the average precision and recall of the algorithm model proposed in this paper on the test dataset are 94.1% and 93.6%, respectively. The values are respectively higher than 93.2% and 91.7% of the original YOLOv13n model algorithm. While improving detection performance, it can also balance the computational efficiency of the algorithm. Therefore, it can realize fast and accurate defect detection of photovoltaic panels, providing technical support for photovoltaic power generation in new energy systems.