Solar panel defect detection based on improved YOLOv8-SDLW

At present, in the field of solar panel surface defect detection, the commonly used detection model has the problems of low detection accuracy, low real-time performance and large number of parameters, etc. In this paper, we propose an algorithm YOLOv8-SDLW based on the improved YOLOv8. In this paper, we propose a solar panel surface defect detection algorithm YOLOv8-SDLW based on the improved YOLOv8, which enhances the model’s ability to perceive the panel’s features at various scales by adopting the SPDConv convolution module, optimises the feature extraction process by combining with the DySample up-sampling module, introduces the LSCD detector header, reduces the number of parameters, and improves the segmentation capability of difficult images by using the WIOUV2 loss function. The WIOUV2 loss function is used to improve the segmentation ability for difficult images. Compared with the original model YOLOv8n, the improved model improves the mAP@ 0.5 by 3.8% and the accuracy by 3.3%, and reduces the number of parameters and computation by 68.1% and 66.9%, respectively, compared with YOLOv8s. It satisfies edge computing or real-time detection scenarios that require both accuracy and lightweight.