Semantic segmentation of remote sensing rooftops based on DAnet-SegFormer

The challenge in estimating the available photovoltaic (PV) capacity in urban areas lies in accurately calculating the roof area suitable for installing PV equipment. Current remote sensing-based building roof identification tasks suffer from issues such as slow recognition speeds and poor recognition accuracy, which significantly hinder the calculation of PV roof resources. To address these challenges, this study proposes the DAnet-SegFormer model, which combines the SegFormer model with DAnet. This model exhibits significant improvements compared to the original model in terms of defect correction. We evaluated the performance of our method through ablation experiments and comparative experiments. The results demonstrate that our approach outperforms popular models in various metrics, improving both the accuracy and generalization capabilities of the model. The method studied in this paper is more conducive to accurately assessing urban PV potential and enhancing energy utilization efficiency.