HYBRID SEMI SUPERVISED MULTIMODAL YOLO11 FRAMEWORK FOR ROBUST SOLAR PHOTOVOLTAIC PANEL DEFECT DETECTION

Abstract

Solar photovoltaic systems have become one of the most important renewable energy technologies for sustainable power generation. However, photovoltaic panel defects such as cracks, hotspots, thick lines, and broken fingers significantly reduce energy conversion efficiency and increase operational maintenance costs. Existing deep learning based photovoltaic defect detection systems still suffer from several limitations including dependence on fully labeled datasets, weak robustness under environmental disturbances, insufficient small defect detection capability, and high computational complexity. To address these challenges, this paper proposes a Hybrid Semi Supervised Multimodal YOLO11 Framework for robust solar photovoltaic panel defect detection under real world environmental conditions. The proposed framework integrates semi supervised pseudo label learning, adaptive multimodal RGB and thermal feature fusion, lightweight YOLO11 optimization, environmental robustness enhancement, and explainable attention visualization within a unified architecture. The semi supervised learning mechanism improves rare defect representation using unlabeled photovoltaic data, while the adaptive multimodal fusion strategy combines structural and thermal information to improve hidden defect localization. Experimental results demonstrate that the proposed framework achieves superior performance compared with existing photovoltaic defect detection methods. The proposed model achieved a precision of 92.8 percent, recall of 90.1 percent, and mean average precision of 93.6 percent while maintaining low parameter complexity and efficient inference speed. Environmental robustness experiments further confirmed stable performance under illumination variation, shadow interference, thermal noise, and dust distortion conditions. The explainable visualization module improved transparency by highlighting the defect regions responsible for model predictions. Overall, the proposed framework provides an accurate, lightweight, interpretable, and deployment efficient solution for intelligent photovoltaic defect detection systems operating in real world environments