基于图像分割的屋顶可用面积提取进行光伏资源评估

Abstract In large-scale roof photovoltaic resource (RPV) assessments, usable roof area extraction has traditionally focused on outer building contours, limiting complex spatial analysis and leading to rough evaluations. This paper introduces a refined method for roof usable area extraction based on both outer and inner contour segmentations. First, the Unet network is employed to segment the roof’s outer contours. Then, a dual-branch encoder network, InSF-TransUnet, based on CNN and Transformer, is proposed. Based on TransUnet, a Multi-scale CNN encoder is incorporated to balance the local and global features. Next, in the decoding phase, multi-scale feature fusion is employed to achieve high-precision segmentation of the inner contours. Finally, the usable area is refined based on the segmentation results, and the roof photovoltaic resources are assessed. The practical feasibility of the algorithm is validated through a case study of a school area.