Detection For Buildings And Their Changes In Remote Sensing Images Based On Deep Learning

Since manually detecting the situation of land resource utilization is expensive and arduous,this article combines the convolutional neural net-work with high-resolution remote sensing images to exploit the value of high-resolution remote sensing images,so that to help the government im-migrate from informationization to intelligentization.It has a great social and economic values.This article uses remote sensing images as the exper-imental objects,to investigate methods of detecting buildings in remote sensing images based on Res-Unet and detecting for building changes in remote sensing images based on FlowS-Unet.First,this article proposes a deep learning framework named Res-Unet to detect buildings in remote sensing images.This framework uses U-Net as basic model and replaces the convolutional layers in U-Net with the residual module in ResNet18.The framework also integrates multi-scale cross training and Adam algorithm to perform the training process.This model enables automatically detect for buildings in remote sensing images.The method of combining deep learning and detection for build-ings can mainly be divided into three procedures:data preprocessing,model training and result prediction.Finally,the post-processing is used to improve the predictions.The experimental results show that using Res-Unet model can detect buildings with high accuracy and good efficiency.Second,the deep learning framework named FlowS-Unet is proposed to detect changed buildings in remote sensing images.This framework ap-plies refinement structure in U-Net,which is inspired by the enhanced re-finement structure in Hypercolumns and FlowNet.Besides,It also inte-grates multi-scale cross training,multiple losses and Adam algorithm to realize change detection for buildings in remote sensing images automati-cally.The method combining deep learning and detection for building changes is mainly divided into three procedures:data preprocessing,model training and result prediction.The post-processing operation is further used to improve the predictions.The experimental results show that using FlowS-Unet model can detect building changes accurately and efficiently which means it is highly valuable.The above results show that the deep learning framework can be ef-fectively applied to the automatic detection for buildings and their changes in remote sensing images,so that it provides an efficient solution to manage land resources.