Research And Application Of Multi-sourse Imaging Target Recognition Based On D-S Evidence Theory

Image fusion theory and its application technology have obtained rapid development of recent years.In many implementations,the combination using of visible light and infrared sensors played key functions and became a worldwide research hotspot.Due to the information fusion processing involves large number of inexact reasoning problems,D-S evidence theory has evolved one of the most important tools for solving knowledge representation and multi-source integrated decision.Aiming at fusion target recognition at decision level based on infrared images and visible light images,the major works in this thesis are:1.Information fusion technology and D-S evidence reasoning theory are introduced briefly.The image features of infrared and visible light images are analyzed.The relevant image feature extraction methods for target fusion recognition are stated.The basic concepts of D-S evidence theory and the reasoning solution are reviewed.The principal problems among its practical application are illustrated.2.Refer to interesting objects' various image features implied by types diversity and essential differences,an extendable approach is proposed to obtain the basic probability assignment in D-S theory oriented on multi-type image features.Based on fuzzy reasoning theory,an effective recognition domain is built according to the target characteristics.The algorithm for determining basic probability assignment is presented by defining fuzzy membership function through image features priori knowledge and deducing corresponding likely-hood function.Thus multi-features image fusion and decision-making can be achieved by classical D-S evidence theory.So the target fusion recognition can be implemented by using the features from infrared and visible images.3.In consideration of the uncertain and unintuitionistic disadvantages of the D-S reasoning in dealing with conflict evidence and the surveys of former methods,the conflictive degree of evidence is presented by vector similarity and evidence distance.Meanwhile,a modified conflict evidence process approach based on compound similarity is proposed.By the average weighted processing for evidences,the problem of highly conflicting evidence fusion is solved to mitigate the interference of robust decision-making.The presented uncertain measurement for conflicting evidence can also be an effective matrix to assess the quality and harmony of information source for multi-sensor management in actual fusion task,etc.The above methods were tested by real data and synthetic data.The algorithms' feasibility and effectiveness were verified preliminary.