Building extraction based on improved message passing network

Abstract

In this paper, an improved building edge extraction method based on message passing networks (MPNNs) is proposed. In the field of high-resolution remote sensing image processing and analysis, the accuracy of building edge extraction is studied from three aspects: data set construction, network training and post-processing of edge probability map. Based on the WHU building data set, the proposed method can effectively extract the building edge, and can maintain high extraction accuracy. PPG Net, L-CNN and UNet++ algorithm is used to compare with the results of this paper. Through the analysis of experiments, the results show that the method proposed in this paper has advantages over PPG Net, L-CNN and UNet++ algorithm in the accuracy, recall and F1 score of region extraction.