Research On Super Resolution Reconstruction Of UAV Image Based On Sparse Representation

In many actual productions,people always hope to obtain clearer UAV images.However,in the actual image collection process,due to the limitations of external factors,imaging systems,sensors and other factors,the UAV image will be degraded,making the obtained data not very ideal.In the three-dimensional model reconstruction and digital orthophoto map Production and other aspects will be greatly affected.Therefore,how to efficiently restore the ideal drone image becomes a very serious problem.Superresolution reconstruction refers to an image processing method that uses multiple frames of low-resolution images to reconstruct a high-resolution image using a certain method.At the practical application level,the super-resolution reconstruction method can obtain relatively high-quality images without improving the hardware performance.Therefore,this article attempts to apply this method to the field of UAV image processing.As the current mainstream method of super-resolution reconstruction,sparse representation reconstruction has the advantages of a small number of samples and high accuracy of reconstructed images.However,the main problems that need to be solved when using this method to process UAV images include three aspects: sparse dictionary Construct and solve the sparse coefficient and super-resolution UAV image reconstruction.In this paper,the method of sparse representation is used for UAV image reconstruction.The main work is as follows:First,Over-complete dual dictionary construction of UAV images based on sparse representation.This paper proposes the idea of introducing simultaneous training of dual dictionaries on the basis of traditional dictionary training.After extracting a large number of samples,the dual dictionaries are merged,and the K-SVD method is used for simultaneous training,so that the high and low resolution dictionaries have the same size image blocks.At the same time,the corresponding sparse coefficients can be obtained,and finally the high-resolution and low-resolution image blocks are registered to obtain an ideal over-complete dictionary.Second,improving the richness of the UAV image over-complete dictionary.In order to achieve better results on the edges,textures and smooth structure of the reconstructed UAV image,the richness of the dictionary is also essential.This paper proposes an adaptive dictionary method.First,the detailed structure of the image is classified according to the image entropy and variance,and then the smoothing,edge and texture structure of the UAV image are processed by corresponding algorithms.The visual effects and data analysis of the UAV image after reconstruction can be obtained.The overcomplete dictionary in this article has better performance and richer details.The experimental results of subjective index exchange(SSIM)and objective index(PSNR)are available.Better embodiment.Third,verifing the reliability and feasibility of this method.The super-resolution reconstruction method is applied to actual production applications such as threedimensional reconstruction,orthophoto map production,and the comparison of experimental data verifies the feasibility of the method in this article in practical applications.Through the comparison of experimental data,the method in this paper can make the image achieve higher accuracy,and obviously improve the performance and quality of the model in the three-dimensional modeling.