MFFN: Multiscale Feature Fusion Network with Pruning for photovoltaic cell anomaly detection

The accurate detection of surface defects on photovoltaic (PV) cell is the pivot of ensuring the stable operation of solar power systems. However, this task remains challenging due to the diversity in defect sizes and the complexity of the background environment. This article proposes a novel deep learning approach for PV cell surface defect detection based on a Multiscale Feature Fusion Network with Pruning (MFFN). To enhance feature representation capabilities, an innovative backbone network architecture referred to as the Feature Fusion Pruning Efficient Multi-Scale Attention (FFPEMA) network is introduced. The dilated convolutional content-aware reassembly of features (DCCARAFE) module is designed to enhance the flexibility of feature extraction. Furthermore, an efficient adaptive feature integration strategy is proposed to strengthen the model's ability to extract and learn features across different defect scales. The experimental results show that the average accuracy of the method proposed in this paper on the PV-elad dataset has increased by 2.5%, and it also performs well in the small object detection task.