Research On Key Technologies Of Building Detection In Remote Sensing Images

The extraction of ground target information from remote sensing images is one of the most important research subjects and has been widely used in military field,civil field and so on.A large part of the ground targets are buildings,especially in urban areas,the ratio can reach about 80%.At the same time,the research of building inspection is of great reference value to other target detection.Therefore,building detection of remote sensing images is of great research significance and practical value.In the aspect of basic theory research,this thesis introduces the main steps of remote sensing image target recognition,the level division of target recognition and resolution,the common features of target recognition,the transformation and invariance of features,and the classifier.The paper focuses on feature extraction and feature classification,gives the commonly used feature extraction methods and classification methods,and introduces the distance metric learning and the related algorithms.In the research of building detection methods,the paper points out the existing problems of building detection algorithms in remote sensing images: the one is ignoring the building shapes when extracting features,which is not conducive to accurately describe the characteristics of buildings;the other is ignoring the similarities between buildings and the differences between buildings and backgrounds when classifying buildings,which will lead to lower classification accuracy.In view of the above problems,shape-specific feature and discriminative learning(SSF+DFC)is combined to detect buildings in this paper.Shape-specific feature(SSF)collect directional gradient data only in homogeneous regions,and feature updating are performed iteratively,then finally the shape and structure of the building can be well described Discriminative feature classification(DFC)obtains the similarity or difference between samples by learning the distance metric matrix M,and can acquire more accurate classification results.Through the experiments on Quick Bird 2 datasets,and compared with other methods,we can draw a conclusion that utilizing shape-specific feature and discriminative learning for building detection can detect the external contour more accurately and can detect more number of buildings.The reason is that shape-specific features collect spectral information and oriented gradients only from homogeneous regions,and can well capture the shapes and structures of buildings;based on discriminative feature classification,similarities between buildings and differences between buildings and backgrounds are acquired to improve the accuracy of building detection.