Building Extraction on SegFormer Model and Post Processing from Remote Sensing Images

Abstract

Most of the traditional methods are based on remote sensing feature classification technology, which uses different classification methods to extract specific feature types, but the traditional classification process suffers from the problems of high threshold of use, cumbersome data processing process, slow recognition speed, and poor migration. Artificial intelligence, especially machine learning and deep learning, is constantly and deeply affecting our daily life and work, and the impact on intelligent extraction of remote sensing images is also very extensive. Classification and automatic extraction of geographic elements of remote sensing images is a popular research direction in the field of remote sensing. Remote sensing image building extraction has an important application value in the field of geographic information, especially in urban planning, resource management, and ecological protection. Deep learning convolutional neural network is used to recognize buildings in remote sensing images, and the current mainstream SegFormer network structure is selected for intelligent binary classification to extract buildings. The intelligent binary classification workflow ranges from data preparation, model construction, model release to application. Intelligent binary classification can intelligently decipher not only buildings, but also single land classes with obvious feature points such as roads and rivers. The development of deep learning algorithms, to a certain extent, to make up for some of the shortcomings of the traditional remote sensing image analysis methods, post-processing of the inference results, it greatly improves the presentation of the results, as well as the accuracy of the results, to provide a better technical basis for the analysis of remote sensing images.