Photovoltaic Array Extraction Algorithm Based on Modified U2Net

The production, transportation, and installation of photovoltaic (PV) modules can lead to component defects. These defects affect the power generation efficiency and can cause local temperature anomalies, leading to hot spots. Collecting infrared images of photovoltaic power stations using drones equipped with gimbals efficiently detect hot spots in PV modules. The extraction of PV arrays is crucial to eliminate the interference of complex scenes in infrared images. Since infrared thermal images only include temperature information without color information, extracting photovoltaic arrays from infrared thermal images is more challenging than visible-light images. Traditional methods rely on manually preset segmentation thresholds and are easily affected by the environment, resulting in low robustness. This paper proposes an improved image segmentation network, Attention-U2Net, based on the U2Net network structure combined with the CBAM attention mechanism. This network enhances the perception of local details in space and the feature extraction capability. Experimental results show that the method proposed in this paper can effectively extract photovoltaic arrays in complex scenes, demonstrating certain advantages over existing approaches.