A Research On Synthetic Aperture Radar Image Change Detection Methods Based On Segmentation Of Statistical Region Merging

Synthetic Aperture Radar image has been widely used in resource monitoring,land planning and target detection in all-weather all-day work.The successful development and launch of China's Gaofen-3 remote sensing satellites with SAR imaging load marks our all-weather monitoring of global marine resources and terrestrial resources.SAR image change detection is widely used in natural disaster monitoring,geographic data update and post-disaster urban reconstruction because it can extract the change information in different time images in the same area.The process of change detection method is generally divided into four stages: image preprocessing,difference image acquisition,difference image analysis and precision evaluation.Among them,the analysis of the difference image is a crucial factor affecting the test results.In this paper,the image segmentation theory is introduced into the field of SAR image change detection,and the multilayer dynamic sorting statistical region merging method is used as the core to obtain the super-pixel segmentation result of the difference image.The cascade segmentation frame and the Markov random field weight optimization algorithm are used to complete the change detection.Specific work is as follows:1)A new method of SAR image change detection based on MDS-SRM hybrid cascade is proposed to solve the problem of high leakage detection of traditional SRM algorithm.Firstly,an MDS-SRM algorithm based on dynamic sorting is proposed to reduce the error of difference image segmentation.Secondly,the multi-channel differential data set is constructed based on the mutual information minimization criterion to improve the algorithm Merging the binding ability;Finally,a cascade segmentation change detection framework is proposed.The first stage uses the SRM algorithm to map the difference image to the super pixel space.The second stage adopts the MDS-SRM algorithm to dynamically merge the super pixels to obtain the convergence segmentation result.The third stage adopts the simplified SRM method to get the final merger Change detection map.The experimental results show that this method can obtain better detection performance than SRM method and current popular method.2)Aiming at the problem of large error of boundary region based on MDS-SRM hybrid cascade detection method,a MRF based on probability feature statistics is introduced toconstruct a PFDSM change detection framework.The two kinds of potential functions based on the HOG weight information and the gray level-gradient co-occurrence matrix weight information are designed.The MDS-SRM detection results are used as the prior probability distribution model of MRF,and the Gibbs distribution and the maximum posteriori probability are used to iterate Changes in the detection results.The experimental results show that the PFDSM detection algorithm can further improve the performance of the change detection.