| Remote sensing image change detection is one of the leading research directions of remote sensing image processing technology.It is a technology to quantitatively analyze regional change information of multiple remote sensing images taken at different times in the same region.With the continuous development of remote sensing image change detection technology in recent years,its research work has been extended from the initial extraction of remote sensing image change features to dynamic monitoring of vegetation coverage,strategic detection of military targets,disaster assessment,and other application fields.At present,remote sensing image change detection methods can be roughly divided into two categories: supervised learning method and unsupervised learning method.Among them,some supervised learning methods require professionals to consume much energy to label training samples and label data manually.Therefore,the research work and practical application of supervised learning methods for remote sensing image change detection are relatively few.However,unsupervised learning avoids the problem of consuming much manpower to annotate data and attracts more attention in the field of remote sensing image change detection.Based on unsupervised learning remote sensing image change detection technology,this paper proposes two change detection methods to solve the efficiency and accuracy of remote sensing image change detection and applies them to aircraft position change detection research.The main research contents of this paper are as follows:(1)Given the problem that the traditional spectral clustering algorithm cannot be processed and analyzed on ordinary computers due to its large amount of calculation,remote sensing image rapid change detection based on connected region marking method and spectral clustering algorithm is proposed.The research ideas of this method are as follows: 1.Input multi-temporal remote sensing image and preprocess the image;Two kinds of difference images are generated by algebraic operation and fused by the non-local mean algorithm.Thirdly,the difference graphs after fusion are extracted by the ROI segmentation window,and then change detection is carried out by spectral clustering algorithm combined with superpixel segmentation method.The proposed method is applied to high-resolution remote sensing images for simulation experiments and compared with standard reference images,and it is found that there is a high consistency and the processing speed is more than a thousand times faster than the traditional method.To a certain extent,this method solves the feasibility problem of using a spectral clustering algorithm to detect changes in remote sensing images on common computers.(2)In order to improve the change detection accuracy of remote sensing images,a highprecision change detection method of remote sensing images based on improved neighborhood ratio difference map and spectral clustering algorithm is proposed on the basis of research content(1)to achieve high-precision change detection of remote sensing images.Research ideas: 1.Image filtering method based on low-rank sparse decomposition and the structural tensor matrix is used in the stage of image filtering;Secondly,a method of constructing neighborhood ratio difference map based on bias distribution is applied in the construction stage of difference map.3.In the stage of difference graph processing,the parallel processing method is adopted for efficient parallel processing of the segmented subgraph;4.Fine processing is carried out on the changing area of the initial detection map in the generation stage.Simulation experiments show that this method performs well in remote sensing image change detection,and the relevant evaluation index is higher than other algorithms,which fully reflects the effectiveness of this method in high-precision image detection.Compared with the results of traditional spectral clustering in 256 G super-large memory,this method also has higher applicability and achieves the purpose of remote sensing image change detection based on spectral clustering algorithm by using a common computer hardware environment.(3)A remote sensing image change detection method based on spectral clustering algorithm is designed for the extended application of aircraft position change detection,realizing the detection of aircraft position change using remote sensing image and marking its position in the original remote sensing image.The research work mainly includes remote sensing image change detection and image recognition based on deep learning.The change detection work of remote sensing image is to use a spectral clustering algorithm to analyze the change characteristics of the difference map generated by multi-temporal remote sensing image through algebraic operation,and then extract the ROI submap combined with connected region labeling algorithm.In the stage of image recognition,deep learning VGG network is used to train the training set to generate a model,and then the ROI submap is compared with the training model for analysis and decision.If the aircraft target with position change is determined,the position is marked in the initial remote sensing image.All the research contents of this paper are based on spectral clustering algorithm,and carry out scientific research in the field of remote sensing image change detection.The research not only solves the problems of slow detection speed and low detection accuracy of remote sensing image change detection method based on spectral clustering algorithm but also designs an extended application based on this method.The research content has not only high research value but also has broad application prospects.However,current studies mainly focus on simple remote sensing images,and future work will detect changes in hyperspectral images. |
