The Research Of The Remote Sensing Image Change Detection Based On Random Fields

Change detection technology based on remote sensing image is to quantitatively analysis and to identify changing features and process in remote sensing images at different times. It makes use of the satellite remote sensing image at different time phases as well as other secondary data to conduct change detection, which is a key technology in land survey, disaster assessment, environmental monitoring, basic geographic dataset updating and etc. and has been widely applied to various public and military applications. In recent years, how to intelligentize the process and to precisely extract changing information has become a crucial research subject. The theoretical research in change detection can promote not only the application of this technology in practice production but also the development of related image processing theories.This dissertation discusses several key problems associated with the remote sensing image cahnge detection based on random fields:how to intelligently determine threshold, and how to fully used changing information of all bands, as well as how to detect by using diverse characteristics in remote sensing image.The concrete contents are as follows:(1) To solve the lacks of segmentation efficiency, stability, universality and accuracy. This paper proposed a change detection algorithm for remote sensing images based on image fusion and adaptive threshold selection. Firstly, an improved image fusion technology was employed in the process of difference image and ratio image of original data to construct fusion image, based on which, a coarse range of change threshold was achieved by an adaptive iterative operation. Then, by analyzing the discrete levels of the image pixels distributed on both sides of the threshold range, get the final threshold range, and thus get a much more optimal one to extract the final change region. The experimental results suggest that the detection accuracy of this proposed method outperforms the traditional change detection methods, and has certain stability and intelligence.(2) The most of traditional change detection methods are adopted to deal with the remote sensing images based on single-band information, the all bands information cannot be absolutely used, so it difficult to detect the complete information. To solve this problem, a multi-band remote sensing image change detection algorithm based on the MRF is proposed in the paper, first, the each band change information can be obtained from traditional ratio change detection method, then the multi-band information is fused by MRF model, at last, the optimal change information can be obtained. Because it is difficult to compute the model parameter of MRF, a EM+MoLC model is used to iterate the all bands information, we can obtain the optimal MRF model and the complete change information. The experimental results suggest that the detection accuracy of this method in the paper, which has certain stability and intelligence outperform the traditional change detection methods.(3) Most of traditional change detection methods conduct computing process based on pure feature, while not comprehensively using all the feature information of remote sensing image, thus cannot be detected changing information completely. To solve this problem, this paper presents a Multispectral change detection method of remote sensing image based on CRF model. This method firstly computes different images based on original image pairs, on which, different features are extracted according to various rules to form feature vectors; then use CRF model to fuse different image feature on feature vectors; finally, through threshold process to get more complete change information. In the paper, we calibrate the proposed method above with appropriate experiments, which proof that, method performed with CRF can effectively integrate diverse image features, to get a comparatively better result than pure feature.The major contents of this dissertation are outlined as follows:(1) This paper proposed a change detection algorithm for remote sensing images based on image fusion and adaptive threshold selection, and the change information of different images be automatically distilled.(2) A multi-band remote sensing image change detection algorithm based on the MRF is proposed in the paper, the multi-band information is fused by MRF model, and the optimal change information can be obtained. (3) This paper presents a Multispectral change detection method of remote sensing image based on CRF model, and use CRF model to fuse different image feature on feature vectors; finally, through threshold process to get more complete change information.