Research On The Detection Of Dim Target Based On Spatial Sparseness

The infrared dim target detection is a key technology in areas of infrared homingmissile, infrared search and track system and early warning system. When the target isfar away from the sensor, it only covers several pixels in the infrared image,and iseasily submerged in strong clutter background and strong noises.So it is great difficultto detect and track the dim target. It is important to propose a new reliable method todetect infrared dim target under the condition of low SNR.Sparse image representation theory comes to a new theory of signal presentationand processing tool after wavelet and multiscale geometric analysis method because itcan describe the signal clearly and concisely.The thesis uses the sparse representationtheory to detect infrared dim target under different background. This thesis focuses onthe morphological characteristics of dim target in infrared image, and trainsovercomplete sparse dictionary with different morphological and differentcharacteristics to characterize the signal, Then dim target of IR is detected by analyzingthe characteristic difference among clutter, noise and target. The content of the thesis isdiscribed as follows:The principle of sparse theory was introduced, and the characteristics of the signalrepresentation were analyzed, the two kinds of dictionary learning methods and sparsesolution methods also were discussed.A method to detect dim target based on multi-scale sparse dictionary was proposed.This chapter expounded the method to construct the multi-scale dictionary, analysisedthe structure component atoms of multi-scale dictionaries. This method had strongerability to suppress the background compared to the wavelet algorithm and theContourlet method. The sparse coefficients of multi-scale dictionaries between thetarget and background were different. The sparse representation coefficients are fittedby the exponential function, and then the target and background would be classified bythe difference of the parameters of the exponential function.A dim target detection algorithm based on adaptive morphological classificationdictionary was presented. An adaptive morphological dictionary was learned andconstructed by the K_SVD algorithm dictionary, and it included more abundant andmore actual atoms to present target signal. Then Gaussian dictionary, Gabor dictionary,Gabor morphological dictionary and the adaptive morphological dictionary were compared in the aspect of representation capability of the dim target form and the abilityof sparse decomposition. The result is that the adaptive dictionary has more diversetarget morphological characteristics, and the ability of morphological characterization isbetter.A dim target detection algorithm based on adaptive morphological classificationdictionary was presented. The adaptive morphology dictionary was divided into thetarget sub-dictionary and background sub-dictionary, which would describe target signaland background clutter, respectively. The sparse coefficient differences mapped on thetarget sub-dictionary between target and background were significant. The sparsecoefficient of target image blocks was very sparse and larger, while that of thebackground blocks wasn’t sparse and was small. We can use suitable threshold canidentify target.