InSAR Image Segmentation Based On Coherence Map Learning

Interferometric Synthetic Aperture Radar(In SAR), which is the extension and further development of Synthetic Aperture Radar(SAR), mainly focuses on obtaining the digital elevation model(DEM). This thesis is concerned on the segmentation of In SAR based on introducing the In SAR technology. The coherence map has an important physical significance. It is not only used to be the evaluation criteria of the phase map, but also has a good separability of different landcovers. In this thesis, we analyze the characteristics of coherence map, which is used to classify the different land covers. Furthermore, we transfer divisibility of the coherence map can classify different land covers to the target SAR image, resulted in a good segmentation. The major contribution of this thesis is as following:A method of the In SAR coherence map segmentation based on the spatial-coherence statistics is proposed. A statistical analysis of homogeneous regions of the coherence map is made. Good classification capability of the coherence map is found by comparing the parameters of mean and variance in different homogeneous regions; Based on the statistical properties of coherence map, we use Bayesian classification which is based on maximum a posterior probability(MAP) to classify the coherence map initially. Then we introduce Markov Random Field(MRF) to increase the neighborhood information, result in it can reduce the number of miscellaneous points. The experiment results show that the characteristics and spatial information of the coherence map can be used to segment In SAR coherence map effectively.A method of In SAR image segmentation based on transferring coherence map is proposed. We use K-means method to segment the target SAR image initially, then each class data learn a dictionary based on K-SVD algorithm according to obtained the initial label. In order to reduce the computational complexity and the randomness of algorithm, we choose a number of samples which are nearest the clustering centers of each class to add the train data to learn dictionary. In the process of the target SAR segmentation, we transfer the dictionary from the coherence map to the target SAR image, the experiments of two real In SAR images show that the segmentation performance has been improved significantly.This work was supported by the National Natural Science Foundation of China(Nos.61003198, 61472306) and the Fundamental Research Funds for the Central Universities(JDYB140508).