SAR Image Change Detection Based On Conditional Random Fields

SAR image Change detection, which aims at identifying changed areas occurred on the Earth’s surface, is a process of making a direct comparison of a pair of SAR images acquired over the same geographical area at different times. Such a process is attracting a growing interest in various applications in the fields of national defense and national economy, such as disaster management, environmental monitoring, urban studies, forest monitoring and battlefield situation analysis. With the development of SAR, SAR image change detection has been widely studied.This dissertation studies the SAR image change detection based on conditional random fields(CRF) model, given the satelliteborne SAR images acquired over the same geographical area at different times. The main contents of this dissertation are summarized as follows:1. A CRF model for SAR image change detection has been proposed in this paper. CRF model is constructed by two parts:the unary potential and the pairwise potential. The unary potential in CRF model is constructed by the support vector machine(SVM) using the texture features which can output the class conditional probability, SVM integrates various features into high-dimensional space, thus handling the non-linear problem and improving accuracy of the model, and the pairwise potential is modeled by the multilevel model which being utilized to regulate the interactions and capture the edge information. CRF directly models the posterior probability as a Gibbs field, and then avoids the problem of model approximations, thus, CRF model has the superiorities of capturing the contextual information of the observed data and complex structures of images through extracting various features. Finally, experimental results on three sets of two temporal SAR images validate the effectiveness of the proposed CRF model.2. In image analysis, the texture features, the spatial interactions and the statistical distribution, play a crucial role in SAR image change detection. To utilize the three kinds of information, we propose a hybrid conditional random fields(HCRF) model for SAR image change detection in this paper. HCRF model is constructed by incorporating the statistics of the log-ratio image derived from the two-temporal SAR images into CRF model. In this way, it is able to integrate the SAR images information, including the texture features of the two-temporal SAR images, the statistics and the spatial interactions of the log-ratio image, into the change detection. And to achieve the integration of the information, HCRF model consists of three parts:the unary potential, the pairwise potential and the data term modeled by the statistics of the log-ratio image. The unary potential and the pairwise potential are constructed in the same way with the CRF model. Generalized Gamma distribution(GΓD) is utilized to model the statistics of the intensity data in the log-ratio image. The parameters in HCRF model are estimated using iterative conditional estimation algorithm, and the iterated conditional modes is used to inference the HCRF model in the process of change detection. Finally, experimental results on three sets of two-temporal SAR images validate the effectiveness of the proposed HCRF model.