Research On Strip Noise Suppression And Learning Based Super-resolution For Millimeter Wave Image

Passive millimeter wave imaging system using objects millimeter-wave radiationenergy differences to achieve imaging, with good penetrating for rain, fog and clothing,and can all-time imaging in a variety of weather conditions, so it has broad applicationprospects in the scene monitoring and security checks. However, due to thecharacteristics of the imaging system, it also has some technical shortcomings. On theone hand, subject to the wavelength and the antenna aperture, the millimeter waveimage obtained tend to have a lower spatial resolution, and the fuzzy extent. To improvethe resolution of the passive millimeter wave imaging by improving the hardware isvery difficult and costly. On the other hand, in order to meet the applicationrequirements, there are often multi-channel imaging system, but was limited by toaccept channel technology and its manufacturing process, making the responseparameters of each channel there are inconsistencies and time-varying, which will resultinto an image the presence of noise bands. In this thesis, the following work which formillimeter wave image stripe noise suppression and super-resolution had carried out:1. Passive millimeter wave imaging theory, the mathematical model of the imagingsystem. Statistical learning theory, analysis of the key factors that affect learningmodel performance.2. The causes of stripe noise for multi-channel system, derivation and establish of themathematical model. On this basis, design stripe noise suppression algorithm usingthe regularization and least squares criterion. and line mean optimization algorithmusing the scene continuity, and its effectiveness is verified by experiment.3. The characteristics of millimeter wave images were studied, and on this basis, apriori model of the millimeter wave image and modeling.4. Study the reconstructed based method super-resolution algorithm, established thepoint spread function of passive millimeter wave imaging system, and designed theblind super-resolution algorithm using the sparse line gradient(SLG) priori model,and its performance is verified by experiment. 5. Research the support vector regression method, analysis the factor which affectedthe super-resolution algorithm performance, and the relationship between learningmodel and computing complexity. Introduced the sparse decomposition into thesuper-resolution, and its performance is verified by experiment.In summary, the algorithm described in this paper, can effectively improve theeffect of passive millimeter wave imaging quality and image processing, with importantacademic significance and excellent value for the research and design of millimeterwave imaging and its image processing algorithms.