Research On Key Technologies Of Rural Building Information Extraction Based On High Resolution Remote Sensing Images

With the acceleration of industrialization and urbanization in China,production factors such as rural labor and land are continuously transferred and accumulating to cities.Due to the reduction of the resident population in countryside,the agricultural sector is weaker and the rural industry fall into deep and prolonged recession.Meanwhile,because of the lack of scientific management for rural land and unified planning for rural construction,the countryside presents the phenomenon of “Expansion of empty”,leading to low ratio of land utility and increase number of vacant and abandoned houses.In order to provide technical support for the comprehensive renovation of hollow villages,rapid access to rural building information has become a research hotspot in the field of remote sensing image processing.There are two main ways,including large-scale field mapping and remote sensing image visual interpretation,to extract buildings information.However,these methods not only cost a lot of manpower and material resources,but also have a long and inefficient graphics period.Therefore,quick access to the information such as the location,distribution,area and damage,from the high-resolution images is significant to field investigation for comprehensive renovation of hollow villages,evaluation for hollow villages and scheme for land consolidation.To address the problems existing in process of extracting building information from hollow village,the dissertation conducts in depth research from three key points.Firstly,for the problem of incomplete extraction of buildings because of occlusion by trees,this dissertation adopts the human-computer interaction method to recover the complete building image by using the structural texture restoration method.Secondly,based on the traditional pixel-based target recognition method,this dissertation improves the threshold selection rules of the color image segmentation method based on Mahalanobis distance,improves the robustness of the algorithm and reduces the computational complexity,thus improving the building recognition efficiency.Finally,the building damage feature vector is constructed according to the color characteristic parameters and texture feature parameters of the building,and the damaged buildings are located and quantitatively analyzed by visualization of damage feature vector.The main research contents and innovative achievements of this dissertation as follows.1.Aiming at the problem of incomplete extraction for rural building in UAV images caused by occlusion from trees,a method of repairing the building image by using the texture-based texture generation algorithm is proposed.The method firstly depicts the occluded boundary and area of the building in the image,and for each occluded area,selects texture sample blocks for texture generation.Due to occlusion,the boundary of the roof or facade of the building cannot be completely displayed.It is necessary to use the boundary to divide the occluded area and screen out the area to be repaired.Then the L-type neighborhood texture similarity repair is performed on the occlusion region by using the texture sample block.Finally,the complete building image with clear boundary is obtained by superimposing the repaired area into the original image.2.Aiming at the efficiency problem of rural building extraction in complex environment,this dissertation proposes an improved Mahalanobis-distance color segmentation method in RGB color space,transforming the complex color discrimination problem of 2D images into a simple spatial relationship of 3D spatial point set.Firstly,the building sample is selected from the image,and the color information of the sample is converted into three-dimensional point coordinates,and filtered in the three-dimensional space to obtain a set of points with low dispersion.Then,filtering the point set in three-dimensional space to obtain a set of points with lower dispersion.Finally,the ellipsoid standard equation is established according to the minimum external ellipsoid of the point set.In the process of recognizing the building,the building information can be quickly and accurately extracted by judging the position relationship between the three-dimensional point corresponding to each pixel on the image and the ellipsoid.3.Aiming at the problems of rural building damage information extraction and damage assessment,in this dissertation,spectral features and texture features are extracted from building in image,and composed of three-dimensional feature vectors.The three-dimensional feature vectors are mapped to RGB color spaces,and the rapid extraction and damage assessment of damaged buildings is realized according to the color space theory.Firstly,the dissertation obtains images of buildings within the experimental area and marks them one by one.Then,the texture features and spectral features are calculated for a single building,and two texture features and one spectral feature that are positively correlated with the damage degree are selected and normalized to a specified interval to form a three-dimensional feature vector.According to the visualization result of the feature vector,information such as the distribution and damage degree of the damaged building is obtained.Quick extraction of damaged buildings in hollow villages from UVA images,can not only locate the damaged building,but also provide preliminary information on the damage of buildings.