Research On Image Super-resolution Reconstruction Based On Micro-scan Imaging

With the development of science and technology,remote sensing technology has begun to develop in a more sophisticated direction.The use of geostationary orbit optical cameras to achieve high-resolution observations on the ground is of great significance to the national economy and people's livelihood.Therefore,the observational resolution of remote sensing satellites has become an important evaluation index for the performance of remote sensing satellites.However,the method to enhance the satellite observation resolution by increasing the optical camera aperture size and the camera focal length,reducing the pixel size and other technical means may inevitably bring about other problems such as the increase of the difficulty and costs of manufacturing and the decrease of the signal-to-noise ratio.In recent decades,with the development of digital image processing,pattern recognition and machine learning technology,many scholars have proposed the use of image post-processing methods to enhance the resolution of the image.The process of using image post-processing to enhance image resolution based on digital image processing technology is called image super-resolution reconstruction.Because of the use of image post-processing,super-resolution reconstruction technology avoids the above problems besides improving image resolution.The image-processing algorithm used is called super-resolution reconstruction algorithm.According to the different number of input image frames processed,the image super-resolution reconstruction algorithms can be divided into multi-frame-based super-resolution algorithm and single frame-based super-resolution algorithm.In this paper,two kinds of algorithms are discussed and compared in detail,and as a main line,a detailed study of micro-scan imaging,sub-pixel technology,multi-frame super-resolution algorithm,and single-frame super-resolution algorithm is performed.And the main work and research results are as follows:(1)Through the analysis of the influencing factors of the optical remote sensor image resolution,some feasible technical solutions for improving the resolution of remote sensor is obtained and discussed one by one.The difficulties of improving the resolution of the optical remote sensor by upgrading the hardware are evaluated.The fundamental of multi-frame super-resolution reconstruction is discussed from the perspective of the theoretical basis of super-resolution reconstruction and the sampling theorem,and the validity of the super-resolution reconstruction algorithm is analyzed.(2)The advantages and disadvantages of various algorithms are analyzed based on simulation of various multi-frame super-resolution algorithms.For the sub-pixel displacement image acquisition problem of super-resolution reconstruction,a corresponding experimental device was designed for experimentation.In addition,this paper proposes an improved maximum a posteriori probability estimation algorithm based on the full variation regularization term for the problem of solving instability in super-resolution process,combined with the momentum update strategy.The algorithm uses the optical flow method for image registration,and takes the full variation regular term as a priori constraint,constructs the minimization of the objective function based on the maximum posterior probability estimation framework,and uses the momentum updating strategy to perform the optimization solution.The simulation and experimental analysis of the algorithm shows that the proposed algorithm outperforms than the previous algorithm.(3)Through extensive investigation and analysis of image evaluation factors,introducing a BRISQUE(Blind reference-less image spatial quality)algorithm without reference image into the super-resolution reconstruction process is proposed for the problem that there is no good evaluation index in practical application for super-resolution reconstruction.It can solves the problem of evaluation of super-resolution reconstruction without reference image.The algorithm is based on supervised SVM(Support Vector Machine)+ SVR(Support Vector Regression)model,and uses SVM to determine the probability of various degenerate factors in the image,and uses SVR to calculate the quality index value of each degenerative type of images.The quality index value is finally weighted according to the probability to obtain the total image quality index.Due to considering all kinds of degradation factors,it is more suitable for measuring the effect of super-resolution reconstruction algorithms.(4)There are many limitations to the technology of multi-frame super-resolution reconstruction,such as the problems of large image registration errors and the difficulty of defining prior knowledge.For the above reasons,a single-frame super-resolution reconstruction algorithm was studied.Simulation analysis shows that the algorithm based on deep learning achieves better performances.However,the algorithm based on deep learning has poor robustness,and image coloring may occur for some images.