Objected-oriented High Resolution Remote Sensing Image Building Extraction Method Research

Building is one of the most important geographic features on the earth and the mark of modern urban development,it is also one of the important goals of remote sensing information extraction which is related to our lives and produce.Because high resolution image is able to reflect the earth feature information with a wide coverage and rapid updating period and it has better analytical ability,its usage in remote sensing extraction as the data source will achieve better performance and obtain the surface objects we need.So we set the object-oriented high resolution image information extraction as the goal,selected building as the subject and conducted a series of experiments.The main work and innovation were as follow.(1)To optimize segmentation efficiency and performance,an improved region growing method concerned about growing principles and growing way was proposed based on original SRG segmentation algorithm.It presented to use high resolution image spectral information to calculate growing principles in region growing segmentation and achieved the image objects from high resolution image.(2)According to building characteristics in high resolution image,the main features of building objects as the basement of object-oriented information extraction,including spectral,texture,geometry features et al.were introduced and proposed a method extracting texture-geometry features of objects.First,the GLCM was improved to obtain texture features to better describe texture information.At the same time,some geometry features were extracted to distinguish buildings from other objects.Finally,8 feature statistical parameters of those features were selected and combined to describe building comprehensively.(3)A method combined of image object texture-geometry features and K-means classification thought was proposed to finish the object-oriented building extraction.It can use image object features and the difference in different object features to analyze and discriminate those features which described the buildings,classify the building objects efficiently and help to obtain the building information we need.In the experiment part,two excel Chinese and foreign high resolution images including GF-2 and Pleiades images were selected and rule-based classification approach was used as comparison algorithm to conduct the object-oriented information extraction.At first,it used the segmentation method in this research to get the image objects.And then the texture-geometry features were calculated and applied in K-means classification.Finally the building information in images was extracted successfully and the object-oriented extraction method proposed in our research was turned out to be valid and feasible through the comparison,analysis and assessment between the experiment and rule-oriented classification results.