The paper presents an improved VarifocalNet for detecting defects in photovoltaic (PV) modules, aiming to enhance both detection speed and accuracy. The key contributions include: 1. **Improved Feature Extraction Network**: A new bottleneck module is designed to replace the first bottleneck module in the last stage convolution group of ResNet-101. This module combines standard and dilated convolutions to increase the receptive field without reducing the output feature map size, improving detection accuracy. 2. **Enhanced Detection Head**: A feature interactor is introduced to enhance feature expression in the classification branch, improving detection accuracy by leveraging dynamic convolution to capture feature relationships between local and global features. 3. **Improved Regression Loss**: An improved GIoU loss function is proposed to better measure the deviation between predicted and ground truth boxes, enhancing detection accuracy. The proposed method is evaluated on the PVEL-AD dataset, showing superior performance in both detection accuracy and speed compared to other methods. The improved VarifocalNet achieves the highest mean Average Precision (mAP) and Recall, while maintaining a faster detection speed than most other methods. The study concludes that the proposed method effectively improves the detection of PV module defects, making it a promising solution for practical applications.The paper presents an improved VarifocalNet for detecting defects in photovoltaic (PV) modules, aiming to enhance both detection speed and accuracy. The key contributions include: 1. **Improved Feature Extraction Network**: A new bottleneck module is designed to replace the first bottleneck module in the last stage convolution group of ResNet-101. This module combines standard and dilated convolutions to increase the receptive field without reducing the output feature map size, improving detection accuracy. 2. **Enhanced Detection Head**: A feature interactor is introduced to enhance feature expression in the classification branch, improving detection accuracy by leveraging dynamic convolution to capture feature relationships between local and global features. 3. **Improved Regression Loss**: An improved GIoU loss function is proposed to better measure the deviation between predicted and ground truth boxes, enhancing detection accuracy. The proposed method is evaluated on the PVEL-AD dataset, showing superior performance in both detection accuracy and speed compared to other methods. The improved VarifocalNet achieves the highest mean Average Precision (mAP) and Recall, while maintaining a faster detection speed than most other methods. The study concludes that the proposed method effectively improves the detection of PV module defects, making it a promising solution for practical applications.