Super-resolution Reconstruction Of Remote Sensing Images Based On Learning

Image super-resolution reconstruction refers to an image processing technology that uses existing digital signal processing methods to increase the resolution of one or more low-resolution images without increasing hardware costs such as imaging equipment.Based on the study of learning-based algorithms,this dissertation proposes improvements to existing problems in existing algorithms.The main research contents are as follows:First,an improved super-resolution reconstruction algorithm for sparse representation remote sensing images is proposed.This method firstly improves the first-order and second-order filter operators of image feature extraction,and the feature information of low-resolution remote sensing images extracted by the improved four one-dimensional filters is more abundant and complete.Then the K-SVD algorithm for dictionary learning is improved.By adding a cyclic dictionary update method,that is,updating the dictionary multiple times in each iteration,the objective function is further minimized.Therefore,the dictionary atom obtained by this method has better performance and accelerates the convergence rate.Secondly,an improved super-resolution reconstruction method for fixed neighborhood regression remote sensing images is proposed.Because the adjusted fixed neighborhood regression image super-resolution reconstruction method treats all neighborhoods in the low-resolution neighborhood in the same way.Therefore,it is not flexible and cannot adapt to the input image block to obtain a better mapping function.Therefore,in order to obtain accurate reconstruction weights and projection matrices,local constrained regression is used to replace the ridge regression in the fixed fixed neighborhood regression.It can well explore the non-linear relationship between low-resolution and high-resolution space and make the regression The solution is more stable.Finally,in Matlab's GUIDE development environment,design and implement a learning-based remote resolution image super-resolution reconstruction system.The system supports local image import,provides image degradation and two super-resolution reconstruction methods for remote sensing images proposed in Chapter 3 and Chapter 4 respectively.It also supports the comparison of images before and after reconstruction,and can save the reconstructed image to the local hard disk.