Building Semantic Segmentation of High-resolution Remote Sensing Image Buildings Based on U-net Network Model Based on Pytorch Framework

Abstract

Modern remote sensing technology has developed rapidly in recent years. The high-resolution remote sensing images brought by new technologies have good application prospects in military and civilian fields, but the information contained in them is also richer, which increases the complexity of remote sensing image analysis and understanding. At present, artificial intelligence technology represented by deep learning has been widely used in the field of image processing. This paper adopts the U-net network model and uses the transfer learning method to train on the remote sensing image dataset published by the French National Institute of Information and Automation (Inria) to verify the effectiveness of the deep learning semantic segmentation method on high-resolution remote sensing images. and stability. Experiments show that the model has an accuracy of 86.86% in extracting buildings from images, a recall rate of 82.54%, and an average intersection ratio of 84.53%, which is effective in semantic segmentation of high-resolution remote sensing images.