The Research Of Key Techniques In Multi-Source Remote Sensing Image Fusion Processing

Multi-source remote sensing image fusion is a frame of method that synthetizes several images of the same environment or the same object. It can acquire information with high qualities, producing more precise, more complete and more reliable estimation or judgement. So the research on multi-source remote sensing image fusion has become hot in remote sensing application fields. Analyzing the advantages and disadvantages of the three levels of image fusion, a technique route of image fusion is formed during the research of our project, which integrates the advantages of all three levels' fusion techniques. And on this basis, the dissertation focuses on the study of the three key techniques of pixel-level registration, feature association and decision-level fusion.In chapter 1, various causes of remote sensing image distortions are expatiated. And on this basis, emphasizes to analyze the existing problems of pixel-level registration, feature association and decision-level identification fusion in image fusion system.In chapter 2, through the studies of image registration theory, rectification techniques are introduced into traditional area-based matching algorithms to overcome the effect of complex structure distortions, which increases its adaptability. The influence of choosing control points on registration performances are investigated for the control-points-based matching algorithms, some rules are summarized which tutor the selection of control-points.In chapter 3, utilizing the actual military targets, the feature-based associating algorithms are simulated, which shows that the feature-based associating algorithms can have better adaptability to imaging distortions.In chapter 4, a method is proposed that combines D-S evidential theory and BP network to identify targets in multi-source remote sensing images, which realizes the application of uncertain reasoning models in multi-source remote sensing image fusion system. It is proved to derive more reliable identification through decision fusion, with a higher recognition rate than that of Bayes method.Chapter 5 gives a brief summary and some prospects for further work.