Building Extraction Of High-resolution Remote Sensing Image Based On Segmentation Method

The buildings as an important object in the High-Resolution Remote Sensing Image(HRRSI),its automatic extraction and interpretation have important applications in economic construction and national defense construction such as new urbanization construction,digital city,military reconnaissance,etc.The research scholars at home and abroad have done many researches on HRRSI processing.In view of this,based on theory of HRRSI building extraction,this thesis presents and realizes several mentalities and methods with practical application value and academic reference value.1.Experiments for the adaptability of traditional segmentation algorithm in HRRSI extraction of the buildings are carried out.In this thesis,four classical object-oriented segmentation algorithms are introduced to extract the buildings,including edge extraction,clustering,regional growth and superpixel.The segmentation results of each algorithm are analyzed visually and quantitatively to draw a conclusion about the adaptability between the algorithm and the HRRSI.2.The problems of low edge fit,complex feature extraction and long running time in the SLIC segmentation algorithm are improved.This thesis proposes an improved SLIC algorithm,which combines the gray features of the building with the position feature to reduce the dimensionality of the feature vector and computational redundancy,and the efficiency of the algorithm is improved,the edge margin has a significant improvement.3.A new method of regional merging for SLIC over-segmentation problem is introduced.This thesis presents a building extraction algorithm in the HRRSI based on improved SLIC and regional adjacency graphs.According to the characteristics of the image,the weight of regional adjacency graphs is proposed,and then the regional merging is carried out.This method can be used in the same image to separate buildings with different structural characteristics and color characteristics,and the segmentation accuracy and operating efficiency are both improved.