Accurate Detection of Solar Panel Defects Using RMN-YOLO With Multi-Scale Edge and Attention Enhancements

Accurate and efficient detection of defects in solar panels is essential for maintaining the stability and reliability of photovoltaic (PV) power generation systems. However, small defect sizes, complex backgrounds, and varying illumination conditions pose significant challenges to conventional detection methods. To address these issues, this study proposes RMN-YOLO, an enhanced detection model based on YOLOv11—the latest real-time object detection framework that integrates multi-scale feature extraction and high-speed inference. RMN-YOLO introduces three key improvements: a Residual Spatial and Channel Reconstruction Convolution (RSCConv) module in the backbone to strengthen spatial–channel feature representation; a Multi-Scale Edge Information Enhancement (MSEE) module in the neck to capture fine-grained defect boundaries; and a Dynamic Head (DyHead) with multi-attention mechanisms to improve small-object recognition and contextual awareness. Experiments on a publicly available solar panel defect dataset demonstrate that RMN-YOLO achieves a precision of 93.3%, recall of 82.7%, and mAP@50 of 90.5%, surpassing mainstream detectors such as SSD, Faster R-CNN, RetinaNet, and multiple YOLO variants. These results verify that RMN-YOLO effectively addresses the challenges of small-target detection and complex backgrounds, offering a practical and reliable solution for intelligent PV defect inspection.