基于深度学习的遥感影像建筑物变化检测方法OA

In the research on building change detection in remote sensing images,issues such as interference from irrelevant targets and insufficient feature interaction between bi-temporal data have long constrained the improvement of detection accu-racy.To address these technical challenges,this paper proposed a deep learning-based building change detection method for remote sensing images.The core of this method lies in the construction of a multi-scale feature fusion mechanism,which consists of three key components:a feature extraction module for multi-level feature acquisition,a computation module for feature difference analysis,and an upsampling module for feature reconstruction.Firstly,a feature extraction scheme was designed that balanced computational efficiency and model simplicity,optimizing the use of computational resources while maintaining feature representation capabilities.Secondly,a spatiotemporal feature collaboration mechanism was formed by integrating temporal feature interaction with spatial feature aggregation,which not only reduced the impact of irrelevant tar-gets but also promoted deep integration of bi-temporal features.Finally,a progressive multi-scale difference feature mask upsampling technique was developed to enhance the reconstruction quality of changing regions.To evaluate the method's performance,experiments were conducted on two publicly available benchmark datasets,large-scale earth vision for image change recognition building change detection dataset(LEVIR-CD)and Wuhan University change detection dataset(WHU-CD).The results show that the proposed method achieves an F1 score of 92.15%on the LEVIR-CD dataset and 90.47%on the WHU-CD dataset,outperforming existing mainstream methods and strongly demonstrating the value and practical appli-cation prospects of this approach.