Research On The Key Technologies Of Multi-sensor Fusion Target Recognition

Withthe more complex environment of modern battlefield,the data attained fromsensors are usually imprecise, partial and unreliable. The technology of multi-sensorinformation fusion synthesizes information from different sensors through mutualassociationandcomplement.Bythiswaytheuncertaintyandrestrictionofsinglesensoris resolved, and the performance of whole system improved, as well as the targetdescribed comprehensivelyandtruly.Since millimeterwave(MMW)and infra-red (IR)are complementary in performance, they are often used in multi-sensor fusion system.For target recognition part of MMW/IR fusion system, we've analysed the fundamentaltheory of multi-sensor fusion in this paper, and then done researches on the keytechnologiesoftarget recognitionbasedonfeature fusionanddecisionfusion.The recognition model of feature fusion includes several processes: featureextraction, feature association and fusion, and classification. In this paper a method offusion target recognition based on KFDA (Kernel Fisher Discriminant Analysis) ispresented. Firstly, joint vectors are constructed from features gained by MMW and IRsensors. Then, KFDA is adopted to obtain nonlinear feature. Finally, an improvedk-nearest neighbor classifier is implemented to recognize targets. Experimental resultswiththerealdatashowthattheproposedmethodisreasonableand effective.For decision-level fusion recognition, a method of MMW/IR fusion for targetrecognition based on gray relational analysis with adaptive weight and D-S theory ispresented. Firstly, the gray relational grade of each sensor is computed and treated asthe basic probability assignment function. Then, the fusion recognition result iscomputed by D-S theory.At the step of grayrelational grade calculation, the algorithmof entropy analysis is adopted to adaptively determine each weight, and as a result thesystem performance is improved. Experimental results show that the proposed methodiseffective.