| Work Detail |
The novel technique is based on the VarifocalNet deep learning object detection framework, which has reportedly been fine-tuned to achieve faster and more accurate results. Compared to other such methods, the new approach was found to be the most accurate and the third fastest. A research group from Beihua University of China and Northeast Electric Power University has developed a novel photovoltaic defect detection method based on deep learning of electroluminescence (EL) images. “Defects in PV cells can lead to module failures, which can result in reduced energy output and pose safety risks to the system,” the researchers said. “Therefore, regular inspection and maintenance of PV modules is essential to ensure maximum performance of the PV system throughout its lifetime.” The proposed method is based on the VarifocalNet deep learning object detection framework, which is a method aimed at accurately classifying a large number of candidate detections in object detection. The ResNet-101 deep convolutional neural network is used as the backbone for feature extraction. To make VarifocalNet faster, the group designed a bottleneck module with smaller parameters, i.e., a layer of the neural network designed to reduce the number of parameters and computational complexity. “To improve the detection accuracy, we design a bottleneck module without reducing the feature map size to replace the first bottleneck module used in the backbone last-stage convolution pool in VarifocalNet,” the scholars explained. “To further improve the detection accuracy, we also design a new feature interactor and improve the regression loss function.” The new detection method was trained and tested on the PVEL-AD dataset, which contains 4,0000 near-infrared images with a range of defects, such as cracks, scratches, black cores, and horizontal dislocations. Additionally, other detection methods were tested on the same database as a reference: RetinaNet, DDH-YOLOv5, Faster GG R-CNN, Cascade R-CNN, the unimproved VarifocalNet, the improved Faster R-CNN, and the improved YOLOv7. In EL images, defects appear as dark gray lines and areas, and the new method is said to detect them quickly and accurately. “Our method exhibits the highest mean precision (mAP) and recall, indicating that the defect detection accuracy of our method is superior to other methods,” the team stated. “Moreover, it also has a faster detection speed than other methods except the DDH-YOLOv5 method and the improved YOLOv7 method.” The researchers highlighted that while their model is a two-stage method, both DDH-YOLOv5 and the improved YOLOv7 belong to the one-stage method. “The two-stage method features a complex network structure, resulting in higher detection accuracy and slower detection speed, while the one-stage method employs a relatively simpler network structure, resulting in faster detection speed and lower detection accuracy,” they further explained. The novel method is described in the paper “ Defect detection of photovoltaic modules based on improved VarifocalNet ”, published in Scientific Reports . |
