Abstract: There are some problems in the infrared images of photovoltaic modules taken by UAVs, such as weak contrast, different defect shapes, and large scale changes. In order to solve the above problems, a defect detection algorithm for photovoltaic modules based on improved YOLOv8 was proposed, firstly, the InceptionNext module was introduced into Backbone and Neck, respectively, to obtain richer context information by decomposing the deep convolution of large kernels, secondly, the space-to-depth (SPD) module was introduced to realize downsampling through feature recombination in the space-to-depth direction, so as to reduce the loss of feature information and strengthen the feature learning ability of small target defects, and finally, Polarized was introduced into Neck self-attention) mechanism, which captures the key feature information on the feature layers of different scales, adaptively adjusts the weights between different channel groups, and enhances the feature representation ability. Experimental results show that the mAP₅₀ and mAP_50:95 of the proposed algorithm reach 79.2% and 44.5%, which are 8.6% and 10% higher than those of the baseline model, respectively, indicating that the proposed algorithm can effectively improve the detection accuracy of infrared image defects of photovoltaic modules.