一种基于深度学习的多源信息融合方法用于高精度光伏识别：U2-Net图像分割与多光谱筛选的集成

Abstract Accurate photovoltaic (PV) identification offers a promising prospect for site selection and wide penetration of future PV systems. This study was dedicated to enhancing the precision of PV identification techniques within urban environments through the integration of the U 2 -Net neural network image segmentation model and the multi-spectral PV screening technique. The U 2 -Net model conducted the image segmentation on visible light satellite images to obtain coordinates and areas of existing PV sites. The multi-spectral screening technique then processed the image segmentation results with multi-spectral satellite images to screen out misidentified samples. The Photovoltaic Index (PVI) and its normalized expression (nPVI) were established to improve the screening performance of the technique. A detailed case study showed that the proposed deep-learning multi-source information fusion method achieved high accuracy PV identification with the intersection over union (IoU) and precision increased by 7.13 % and 8.66 %, respectively, reaching 91.37 % and 93.86 %. This was because the false positive samples of the U 2 -Net image segmentation model were effectively filtered out by the multi-spectral screening technique. The PV deployments in the three megacities (i.e., Guangzhou, Shenzhen, and Dongguan) of Guangdong Province were identified using the developed method. The results showed that the distribution of PV samples is denser in areas with a higher level of industrial development and land resources, while in urban core areas with high urbanization density and limited spatial resources, the distribution of PV samples is sparser. In conclusion, the proposed method has the potential to enable accurate PV identification by efficiently harnessing multi-spectral data. It can assist in formulating more targeted PV deployment strategies, guiding the rational allocation of PV installation resources in various urban contexts, and promoting the integration of PV systems with urban planning, thereby contributing to the global advancement of renewable energy development and the realization of sustainable urban energy transitions.