Research On SAR/Infrared Image Fusion And Target Detection Based On Sparse Representation

Synthetic aperture radar imaging guidance can provide high resolution SAR images at any timeand any weather. Infrared imaging guidance can recognize camouflage of the targets by capturinginfrared radiation. Research on SAR and infrared image fusion is meaningful for improving targetdetection and precision guidance ability under complex background. Based on compressed sensing(CS) theory, this paper approaches SAR and infrared image fusion and target detection algorithms viasparse representation. The main work and contribution are as follows:Firstly, this paper made a research on sparse solving algorithms. A search path improving methodhas been proposed in gradient pursuit (GP) to enhance robustness and accuracy. In other compressivefusion algorithms, the fusion results were affected by the sparsity of input images which were hard toestimate. Thus, for compressive measurements of SAR and infrared images, a new fusion algorithm ofstandard deviation weighted averaging was proposed. The fusion image was then reconstructed by thestep improved GP method. The experiments showed that the proposed fusion algorithm can achievebetter performance without prior information of the input images.Secondly, this paper made a research on dictionary. The fixed dictionary couldn’t adjust atoms todifferent input images, thus an over-complete dictionary construction method based on multi-sourceimage samples has been proposed. The dictionary was building according to the sparse structure ofSAR/infrared image. For the sparse coefficients, a fusion method of energy weighted averaging wasproposed. The experiments showed that, comparing with other compressive methods, the proposedalgorithm can obtain better sparse representation of the images and be more applicable.Finally, this paper made a research on application of CS theory in target detection. In order toweaken the influence of background clutter, two detection methods have been designed separately,one was on the model of sparse point scattering in SAR image, the other was by constructingover-complete target dictionary in infrared image. In order to take full advantage of target information,union membership function was established by combining fuzzy theory and the above methods. Thena target detection algorithm on the basis of soft decision-level fusion was proposed. The experimentsshowed that the proposed method can weaken the influence of clutter effectively and improve thedetecting efficiency.