Research On Extraction Technology Of Suburban Buildings Based On Multi-level Visual Context Information

With the continuous improvement of remote sensing data acquisition capability,there is an urgent need in the field of civil,military and anti-terrorism for the detection technology of suburban buildings based on large-scale remote sensing images acquired by airborne and satellite platforms.However,the development of remote sensing image interpretation technology is lagging behind.Researchers at home and abroad have made unremitting efforts to solve this outstanding contradiction.In suburban remote sensing images,buildings have different shapes and large intra-class differences,which lead to limited detection performance and slow calculation speed of existing methods.In order to solve the above problems,this paper proposes a detection algorithm for remote sensing buildings in suburbs based on multi-level visual context information,focusing on key technologies such as extraction of suspected areas and target identification in remote sensing target interpretation,and constructs an overall detection and identification algorithm framework based on this.The main work of this paper can be summarized as follows:Firstly,a fast screening algorithm for suspected areas of buildings based on geometric space graph theory is proposed.This algorithm obtains the rough segmentation result of the suspected area of the building by using the gray features of the remote sensing image with large field of view in the suburbs and the graph theory segmentation method.According to the geometric characteristics of the building itself,the false alarm is targeted to eliminate,so as to ensure the accuracy of the whole model.At the same time,a fast and precise segmentation method based on block size is proposed to find a more precise segmentation threshold,so as to obtain the precise segmented suspected building area.Secondly,this paper constructs a unified framework of multi-level context information coding fusion algorithm.In order to make full use of the rich contextual information in remote sensing images,this algorithm is proposed.At the same time,it avoids the intra-class differences between suburban buildings and ensures the accuracy of identification and confirmation.The context relations among target area,target circumference scene and target and circumference scene are constructed.The fusion application of three kinds of context information is realized after visual bag feature coding representation.Thirdly,this paper designs a suspicious region identification method based on context semantic web.The first layer of the classification network describes the middle-level semantic features of typical parts of the suspected building area,and the second layer of the classification network uses the middle-level semantic features to identify the suspected building area,effectively eliminating false alarms.By using this method,the suspected areas of buildings can be accurately identified and classified,that is,the suburban building areas can be obtained and the non-building suspected areas can be multi-classified.Finally,on the basis of the above three key technologies,this paper constructs a complete suburban building detection software system.It realizes fast and accurate detection of suburban buildings in large-scale remote sensing images and accurate classification of several types of false alarms.Supported by a large number of optical remote sensing image data,this paper uses multi-level and multi-angle experimental means to verify and analyze the efficiency and validity of the key technologies and the overall algorithm framework.