Research Of Building Extraction And Chronological Classification In Southern Jiangsu Province Based On Deep Learning

With the advancement of technology and the advent of the era of big data,there are more and more ways to obtain high-resolution remote sensing images,such as Google Earth,DigitalGlobe,USGS,and geospatial data cloud.As an important feature in remote sensing images of cities and towns,buildings play an important role in the fields of surveying and mapping and urban planning.At present,there are two main methods for building buildings based on remote sensing images: manual manual extraction and algorithm extraction.In the study of rural planning and spatial evolution,the age information of buildings has important reference significance for the development and evolution of villages.With the support of the mentor Sunan project,this paper conducted field research and building photography on three natural villages in southern Jiangsu,and then downloaded 50 images of 0.61 m remote sensing images in the survey area through Google Earth.In recent years,with the rapid development of artificial intelligence technology,its application research in the field of GIS is more and more.In this paper,the deep learning technology in artificial intelligence is used to automatically extract the buildings in the remote sensing image,and the automatic identification of the building age is realized based on the collected photo data of the building.The specific contents are as follows:(1)Based on U-Net neural network,I improved it on the segmentation result and loss function.By using 35 remote sensing images with spatial resolution of 0.61 m for training and 15 samples for testing,the building segmentation model is obtained.Through the comparison of index evaluation and segmentation results,the segmentation result similarity of the model is 79.33%,which can accurately identify the buildings in the remote sensing image,but it is not ideal in the recognition effect of the edge part of the building.(2)Through field research and data review,it is determined that the building age of this paper is divided into: before the reform and opening up,in the 1980 s,1990s and after,and then the characteristics of the buildings in these three years were analyzed and to sum up.(3)Using the classic four models of deep learning(AlexNet,VGG-16,ResNet-50,DenseNet-121)to perform age classification training on the photographs of the building facades.After comparing and analyzing the model results and combining with the actual situation of holding the data,this paper finally chooses AlexNet as the building age classification model,which has a classification accuracy of 95% on the test set,and the overall prediction ability is better,but the ability to judge buildings with vague characteristics is weak.(4)Embed the trained model into the GIS system based on C# + ArcGIS Engine secondary development,develop a simple and practical remote sensing image building automatic extraction and building age auxiliary classification system,realize one of AI in GIS field Applications.