Comprehensive Analysis Of Multi-temporal Remote Sensing Images’ Change Based On Dynamic Bayesian Network

Remote sensing images are widely applied in many fields such as military and industry. The satellite can provide multi-temporal remote sensing images for researchers to detect a certain region. Traditional remote sensing change detection algorithms only generate change-detection map and the types of the change, but they are unable to provide comprehensive analysis and further understanding of the detected changes.It’s obviously that the information of the images are not made full use of. Therefore, it’s important to propose a method to satisfy the need to comprehensive analysis the change of a region which not only can be implemented in the engineering but also has theoretical research value.Aiming to assess regional development, we develop a comprehensive analysis method for human-driven environmental change by multi-temporal remote sensing images. And the data of experiments are the images of two kind of harbor. The method is based on DBN(Dynamic Bayesian Network), and the input parameters are got from change detection and target identification technique. First of all, we introduce the detection method of ships near the pier and oil depot, the detection method of ships is based on Hough transform and region growing, and the oil depot is based on top surface circle or ellipse detection, the ellipse. We also give three traditional change detection method and a method that applied to high resolution remote sensing image by fusing multi-characteristic information.Then, the study about DBN are given, including the history of DBN, the theory of DBN and the application of DBN. In order to adapt to analyze the time-varying multiple changed objects, an observed object-specified Dynamic Bayesian Network(i.e.,OOS-DBN) is firstly proposed to adjust DBN structure and variables. Using semantic analysis for the relationship between multiple changed objects and regional development, all levels of situations and evidences(i.e., detected attributes of changed objects) are extracted as hidden variables and observed variables and input to OOS-DBN. Furthermore, conditional probabilities are computed by levels and time slices in OOS-DBN, resulting in the comprehensive analysis results. Furthermore, some types of input data errors are given to prove the robustness of the method.A civil port and a navel harbor are chosen to be apply data to verify the method researched at the end. The civil port is Huludao new port and we evaluate it’s development from 2003 to 2014. Detect the change of the objects of each two adjacent images which are got from Google earth, and then analyze them by OOS-DBN. The navel harbor is Norfolk navel harbor, and it’s state of military from 2009-2014 is analyzed.