Object Change Detection And Recognition Algorithm For The Object In Satellite Image Based On Object Segmentation

Based on multi-source remote sensing data,remote sensing change detection technology is a kind of computer image processing technology,which can detect,recognize and analyze the variation of object phenomenon,process and state in time domain with the auxiliary of the knowledge database.Generally,it include the multisource data acquisition,the original data preprocessing,change information extraction and change properties determination,information processing and access of precision.The main aim is to determine whether a target change,determine the area of change,identify the category of change,the time and space distribution of the evaluation model.With the development of machine learning and deep learning,deep learning algorithm with big data as the core has been extended from traditional natural image processing to remote sensing image processing.In this thesis,machine learning and deep learming algorithms are combined to carry out the research on object change detection and recognition algorithm in satellite images based on object segmentation,which is of great significance and has a potential of application.With the combination of machine learning algorithm and deep learning algorithm,this thesis focuses on change detection and recognition of the architecture on the satellite image.The main contents and innovation is described as the following:.(1)By analyzing the existing remote sensing image data set,the change detection and recognition of the object in the existing data set can be manifested.In view of the change detection of remote sensing images including the characteristics of before and after time phase images,the former time phase is the reference and thus the latter time phase image is defined.Aiming at the problems of unbalanced distribution of different sample database and too close distance between classes,a sample database equalization module is proposed.In order to solve the problem of insufficient sample size,a data amplification algorithm for satellite images is proposed.(2)In view of the difficulty in identifying building changes from high-resolution remote sensing images,a building change detection method based on stack noise reduction self-encoder(SDAE)is proposed.Firstly,for the matching errors of remote sensing images,SIFT algorithm is used to extract the feature points of different scenes for image alignment.Then,stack noise reduction self-encoder is used to extract image features,and FCM clustering method is used to obtain the changed region.This algorithm exhibits high detection efficiency and can adapt to the spectral differences of different source images.(3)Based on the U-net model,a new model called Wide net(W-net)is proposed.The mixed loss function is used to solve the problem of unbalanced training data for the positive sample and negative sample distribution caused by the cluster distribution and scattered distribution of buildings.Two u-net models are connected and named w-net.The first u-net outputs auxiliary information,such as building topology and pixel distance.The second u-net generates a building mask by dividing each pixel into buildings or non-buildings.(4)Since remote sensing image scene has large information,complicated background,it is necessary for developing the algorithm,which has high robustness,high detection efficiency,and the ability of overcoming the spectral differences between different source images.Therefore,the method combining the above SDAE and FCM is put forward to detect the change area,and then to identify the buildings by using W-net network.Numerical results show that the proposed method according to the structure of the change detection algorithm has high reliability,fast inspection speed.