Research On Building Change Monitoring Along Railway Based On Remote Sensing Image

With the development of railway industry,railway plays an increasingly important role in social economy.Railway construction can promote the rapid development and change of all aspects of the local area,among which,the use change of construction land is the most obvious embodiment.The number and distribution of buildings before and after railway construction will change to a certain extent.Building information,as an important reference factor of land use statistics and urban development,can be obtained in time before and after the railway construction,which can play a reference role in the decision-making of land use and urban planning.In addition,the development of remote sensing technology makes the monitoring of ground feature information develop to the direction of high accuracy,real-time and low cost.It is of great significance to use remote sensing image to monitor the changes of buildings along the railway,whether in image interpretation method or in practical application.Based on the remote sensing image data,this research studies the change monitoring technology of buildings along the railway,and takes the building image along the railway as the data support.The research work of this paper is as follows:First of all,the u-net neural network algorithm is applied to the remote sensing image building change detection task,and the data set is constructed according to the network structure and detection task to ensure a good network model after training.The defects of U-net neural network structure are improved as follows: Aiming at the problem that the u-net neural network can not accurately detect the buildings in the remote sensing image with complex features,this study proposes the improvement of low-dimensional feature enhancement,which can reduce the loss of low-dimensional feature information in the backward propagation process of the improved network,so as to improve the accuracy of building edge extraction.Double constraint function and normalization operation are added to the network to ensure the efficiency of the convergence process of network parameters.It improves the ability of the model to obtain the details of the building and ensures the accuracy of the primary results of the building detection.Because of the shadow or occlusion in the building,the network can not recognize its features well,and the noise can also affect the accuracy of image detection results.An improved super-pixel segmentation algorithm is proposed to optimize the results of building change detection.A method of randomly selecting multiple points and then comparing them is proposed instead of simple uniform distribution method.After each iteration,a filtering operation is introduced to remove the pixels with large difference in color space between the super pixel and the clustering center,and the remaining pixels are used to update theclustering center.The result of super-pixel segmentation is optimized.In this study,a new method of building change detection based on u-net and super-pixel segmentation is proposed.By fusing the improved u-net detection results with the corresponding super-pixel segmentation results,the influence of the shadow,occlusion and local noise on the detection results can be reduced.The high-resolution remote sensing image of the test area is used to detect the changes of buildings in the station area and along the railway track.The method proposed in this study can accurately detect the changes of buildings along the railway,and effectively provide reference for land use and urban planning.