Application of RF-DETR and YOLOv12 Model on a Novel Solar Panel Dataset for Defect Detection in Sustainable Energy System

Solar panels are essential for sustainable energy as they convert sunlight into clean, renewable electricity with minimal environmental impact. However, defects such as cracks and hot spots can diminish performance by causing power loss and accelerating solar cell degradation. Therefore, precise defect detection is critical for effective maintenance and sustainability. This study assesses two innovative models, RF-DETR and YOLOv12, using a novel dataset to compare precision, recall, and mean Average Precision (mAP) with the aim of enhancing defect detection accuracy and efficiency. A four-stage methodology is employed, wherein high-resolution drone images from thermal and high-definition cameras are processed to improve quality. Noise reduction and contrast enhancement are applied prior to model implementation to ensure precise and reliable defect detection under various conditions, thereby evaluating model effectiveness and accuracy. RF-DETR demonstrated superior performance over YOLOv12 across all metrics, achieving a precision of 0.83, recall of 0.78, and F1-score of 0.81 in single-class detection, compared to YOLOv12x's precision of 0.66 and F1-score of 0.72 versus 0.57 in multi-class tasks.