Research On Change Detection Method Of Remote Sensing Image Based On Variable Weight Markov Model

Remote sensing image change detection technology is a hot research in the field of remote sensing image processing,the technology is mainly for remote sensing image analysis,extract the changes over time on the image information,to determine the surface or feature corresponding to the change of area,to get information of interest,so that the follow-up work of the research and application.Remote sensing image change detection technology has been widely used in various fields,such as urban development planning,soil and water environment investigation and disaster monitoring,military investigation and other fields.In recent years,a large number of researchers in the field of remote sensing continue to make research and innovation in the accuracy and reliability of remote sensing image change detection technology.However,there is still no algorithm that is suitable for change detection of two-phase remote sensing images obtained by various sensors and various shooting conditions.Based on this,this paper makes an in-depth study of the current research status and deficiencies of existing remote sensing image change detection methods,and realizes remote sensing image change detection based on Markov model.The concrete research works and contents are segmented into the following two sections:(1)Most of current remote sensing image change detection algorithm for image pixels,grey value,the lack of spatial information of image pixels using Markov model has a strong spatial expression ability and the theory of system,but in the actual process of Markov random field modeling,tag field of energy and the characteristics of the field of energy in the proportion of the energy function is the same,the segmentation edge in the change detection results has the defect of blurring,which leads to the low accuracy and accuracy of change detection.Based on the theory of bayes rule,the image segmentation problem can be based on maximum a posteriori probability integral framework into the problem of energy function minimization value,this paper introduced variable weight of ideas,set up from a weighting function to adjust the space field energy field energy proportion size and spectral characteristics,finally using iterative algorithm(ICM)condition optimized segmentation result,obtain the final change detection results.Through comparative experiments on two groups of real remote sensing images,the accuracy of the proposed method is compared with that of K-means clustering algorithm and MRF algorithm.The accuracy of change detection results was improved.(2)The change detection of remote sensing images based on Markov model takes into account both the grayscale information of pixels and the spatial relationship of pixels,but the consideration of spatial relationship among pixels is not reasonable enough,leading to low precision of change detection.In view of the above situation,this paper proposes a variable weight MRF remote sensing image change detection method considering pixel space gravity.First through change vector analysis(CVA)for two phase difference of the initial image,then USES the fuzzy c-means clustering(FCM)algorithm for remote sensing image initial fuzzy segmentation result,use of the space gravity model to pixel membership information and the distance between neighborhood pixels is introduced into the Potts model,the characterization of pixels classification mark again,so as to enhance the image detail retention ability;The spectral feature model was established by Gaussian mixture model on the spectral information of pixels.Based on the adaptive value function adjusting the proportion of the energy of spatial marker field and spectral feature field,the optimization iteration was carried out to obtain the final detection result of remote sensing image change.For purpose of verifying the effectiveness of the method,correlational experimental works were carried out on three groups of high-resolution remote sensing images with different sensors and different resolutions,and the experimental results were compared with the existing algorithm.The proposed algorithm showed better edge detection and regional consistency,with better detection accuracy.