Building Extraction From Remote Sensing Image Based On Encoding And Decoding Deep Learning Model

With the development of remote sensing technology,the resolution of remote sensing image has been improved,and the information of buildings like texture and shape becomes clearer,meanwhile,the ground object information in remote sensing image becomes more and more diverse.Therefore,how to extract buildings from remote sensing images more accurately and quickly has become one of the research hotspots.Traditional building extraction algorithms from remote sensing images are no longer appropriate in today’s age of rapid information change,because of time-consuming and labor-intensive.Building extraction from remote sensing Image based on deep learning has become one of the mainstream methods of building extraction from remote sensing images,because it has many characteristics like high automation and high precision.Building extraction from remote sensing image based on encoder-decoder deep learning model has shown great advantage,but most of the building results extracted by algorithms have some problems such as missing information inside the building and incomplete edge information,for solving those problems,this paper proposed building extraction method of remote sensing image based on AGR2U-Net model.In this method,the AG module can make use of the multi-scale feature information obtained from the feature extraction of R2U-NET model to enhance the expression ability of building features,which can improve the sensitivity of model and the accuracy of building extraction.AGR2U-Net model was verified with other methods by using WHU Satellite dataset I and WHU Aerial imagery dataset.The extraction results of the buildings datasets showed that the mean values of the proposed method are the highest in the three evaluation indexes of intersection ratio,pixel accuracy,average pixel accuracy and Recall,respectively.It can be seen that the results are better than those of the U-Net,Improved UNet,Seg U-Net and R2U-NET methods.Moreover,this method can extract more accurate edges in the building extraction from remote sensing images and more complete internal information of buildings with lower missing rates.Due to the diverse terrain information in remote sensing image building,there are many buildings extraction algorithm are susceptible to interference problems such as feature extraction,in order to solve the problem,this paper proposed building extraction method of remote sensing image based on BTS-Net model,this method is a makes full use of the information in remote sensing image to strengthen the expression of building characteristics,and which is a encoder-decoder deep learning model.In this method,BOTNet-50 is used to enhance feature extraction,and SE-Net is used to add building feature channel information to enhance the expression of building features.The proposed method was verified with other methods by using Massachusetts dataset,Deep Lap V3,UNet Improved U-Net and Seg U-Net methods were selected for experimental comparative analysis with BTS-Net.In the training process,BTS-Net performed well and the training accuracy was higher than that of the comparison method.It can be seen that the BTS-Net experiment results is more sensitive to the recognition of building features in the remote sensing image building extraction task.Compared with its comparison method,the experimental results have higher accuracy and can accurately extract building features from chaotic remote sensing image information.