Building Change Detection And Semantic Recognition Based On DSM Multi-scale Sampling

Change detection is a method of processing and comparing the front and back time phase data of the same area to accurately obtain the change information of the features in the area.Buildings are important signs of cities,and their distribution patterns are special,which is of great significance to the detection of changes in buildings in urban areas.Although the structure of DSM data is relatively simple,it contains rich information and can truly show the changes in urban spatial morphology.Therefore,by analyzing the changes of DSM data in different time phases,building change detection can be accurately and efficiently realized.This paper uses multi-scale down-sampling DSM data to detect changes in buildings.The building elements of urban planning and construction are used as the monitoring objects to realize the detection of changes in large-scale urban buildings,which can quickly and effectively extract the map spots of the change area.The test results show that compared with the traditional manual visual interpretation,this method not only has high efficiency,but also guarantees a higher accuracy rate and greatly reduces manual intervention.It is suitable for large-scale DSM building change detection and is DSM data.Building change detection provides a feasible technical method.This paper adopts two periods of DSM data in the same region,and the main work includes the following aspects.(1)A building change detection method based on DSM multi-scale sampling is proposed.The method first calculates the pixel-by-pixel difference of the two periods of DSM data,and extracts the change information through the difference in elevation information.The change information extracted by the difference DSM image is susceptible to the influence of DSM elevation accuracy and noise in the front and back phases.Multi-scale downsampling is performed based on the difference DSM image,the optimal scale is selected,and the downsampling scale spatial sequence image is obtained,which can effectively reduce the The noise information generated when the pixel difference value.In order to accurately detect the changes of buildings,a method of de-forgery analysis based on small area removal and image complexity is used to remove the pseudo-change areas.(2)In order to solve the problems of over segmentation and adhesion caused by direct watershed segmentation,this paper improves the existing watershed algorithm,and proposes the image preprocessing by means drift filtering algorithm.Then the watershed segmentation image is combined with similar regions.The improved watershed segmentation method is used to segment the change,which can effectively suppress the over segmentation and adhesion phenomenon caused by the traditional watershed algorithm,and determine several candidate objects for change,and obtain the ideal change segmentation area.At this time,the candidate change objects contain the information of buildings and vegetation.(3)Combining building vector data for building change detection,projecting DSM candidate change objects onto the vector,mixing and overlaying raster vectors,and then using the vector building information to eliminate vegetation,ground attachments and other non-buildings Change area.Perform classification and extraction of change patterns and semantic recognition of change types to determine the types of changes in buildings,and divide the change categories into three categories: new construction,demolition and unchanged.There are 46 pictures,3 tables and 51 references.