Research on Hot Spot Defect Detection in Photovoltaic Modules Based on UAV Remote Sensing Technology and Deep Learning

The increasing deployment of photovoltaic (PV) systems necessitates advanced methodologies for defect detection to ensure operational efficiency and longevity. Traditional inspection techniques often fall short in scalability and precision, particularly for localized anomalies such as hot spot defects. These limitations highlight the need for innovative solutions rooted in computer vision and theoretical computer science. Conventional approaches, including manual inspections and basic image processing, lack the adaptability and computational efficiency for real-time analysis, leading to delayed maintenance and increased costs. They also struggle with accurate localization of defects, critical for effective remediation. To address these challenges, we propose a framework integrating Unmanned Aerial Vehicle (UAV) remote sensing with deep learning techniques. Our solution comprises two core components: the Neural-Symbolic Defect Mapper (NSDM) and the Dynamic Routing and Reweigh (DRR) strategy. The NSDM leverages a hybrid architecture combining spatial attentiveness with symbolic abstraction, enabling precise localization through a dual-phase pipeline. It adapts across defect granularities and remains effective under weak supervision or noisy labels. Meanwhile, the DRR strategy introduces an adaptive inference mechanism that dynamically switches between a lightweight detector and a high-resolution refinement stage, based on the input’s statistical characteristics. This enhances detection fidelity while optimizing computational resources, achieving real-time performance. Experimental results show that our integrated approach outperforms traditional methods in detection accuracy and runtime efficiency, marking a significant advancement for PV system defect detection.