Abstract: Hotspot defects in photovoltaic (PV) modules directly lead to low power generation efficiency in PV power stations and can even cause fires. To address the problem of the low detection accuracy of hotspot defects in PV modules, an improved faster R-CNN method for hotspot defect detection in PV modules is proposed. First, based on the faster R-CNN object detection model, ResNet101 and enhanced feature pyramid network (EFPN) are introduced to replace VGG16, which enhances the detection accuracy of small target defects. Second, global average pooling is used to replace the fully connected layers, thus reducing the number of parameters required for computation in the faster R-CNN model. Finally, a thermal restart cosine annealing strategy is adopted to update the learning rate and improve the convergence speed of the model during training. Experimental verification and comparison with other models show that the improved faster R-CNN model achieves an accuracy rate of 94.8% for hotspot defect detection in PV modules. The results indicate that the improved faster R-CNN exhibits excellent practicality and accuracy for hotspot defect detection in PV modules, thereby outperforming other models such as YOLOv5 and SSD.