Research On Building Extraction Method Based On High-resolution Remote Sensing Image

With the continuous launch of remote sensing satellites,the spatial resolution of remote sensing satellite sensors has reached the sub-meter level,and more and more valuable information can subsequently be retrieved from remote sensing images.Buildings are places where humans live and are closely related to our lives.Building extraction based on high resolution remote sensing images has always been a research hotspot.Since the launch of the first remote sensing satellite,a lot of research on building extraction have been conducted.Low-resolution images are usually employed to extract built-up areas,whose extracting algorithms are relatively mature and highly accurate,but the algorithms cannot be directly and effectively applied to high-resolution images for building extraction.this thesis started with summarizing previous research on building extraction based on high-resolution images and subsequently developed a new building extraction method with high accuracy and integrity,which employed machine learning method(SVM)by incorporating object-oriented local binary pattern feature and multiple building features.The major contents of this thesis are as follows:Firstly,The processing of high-resolution remotely sensed imagery by using histogram equalization and following bilateral filtering to smooth the interior area and background details of the building without loss of the boundary information;A low-density feature map is applied the local binary mode algorithm,then,images is Segmented based on the mean value shifting method for objects mapping;Secondly,on the basis of the obtained feature objects,the feature map is converted by the principle of rotation invariance and each object is traversed using a sliding window.The magnitude of each mode pair of each object is counted and its LBP characteristics are obtained.In addition,multi-dimensional feature vectors are formed by introducing shadow index,rectangularity,and brightness features.Finally,the multi-dimensional feature vector is input into the support vector machine to complete the building extraction,and the morphological optimization of the building results is used to improve the integrity of the contour.Thirdly,a comprehensive experiment is designed and performed to analyze the effect of the algorithm from the perspective of processioning effect,parameter setting,comparison with other algorithms,and large-scale extraction.Finally,the content of this thesis and point out the problems and deficiencies is summarized in this thesis as well as the plans and solution to this problem.