Research On Hyperspectral And Multispectral Image Fusion Based On Sparse Representation

Hyperspectral imaging technology is mainly used to obtain image data of the same scene in multiple narrow wavelength bands.It is capable of detecting the spatial structure and spectral reflection information of target objects,and thus is widely used in remote sensing monitoring,medical image analysis and food safety.Since hyperspectral imaging systems operate in narrow wavelength bands,the sensors usually need to collect photons in a larger spatial range in order to ensure sufficient signal-to-noise ratio,which results in a much lower spatial resolution of hyperspectral images than multispectral images.How to effectively fuse hyperspectral image data with multispectral images with high spatial resolution in the same scene and recover the reconstruction to get a data cube with high spatial resolution and hyperspectral images is a valuable research direction.Through the analysis of the status of hyperspectral and multispectral image fusion at home and abroad,this thesis proposes a hyperspectral and multispectral image fusion algorithm based on homogeneous superpixel sparse model and a hyperspectral and multispectral image fusion algorithm based on global structure and spectral self-similarity constraint.The main innovative work of the thesis is as follows:(1)For the problem that the traditional superpixel solving sparse representation model does not possess global structural sparsity,a hyperspectral and multispectral image fusion algorithm based on homogeneous supeipixel sparse model is proposed.The method first uses an online dictionary learning algorithm to train from hyperspectral image data to obtain a hyperspectral dictionary and downscale it to generate a multispectral dictionary;considering that the same substances in the scene usually have approximate spectral reflection coefficients,we perform hyperpixel segmentation on the multispectral image,calculate the statistical probability distribution of each hyperpixel,and measure the distance of the statistical probability distribution between the hyperpixels,and merge the hyperpixels whose distance is less than a threshold.Then,we use the structured sparse coding technique to solve the sparse coding of homogeneous superpixels by combining the multispectral dictionary;finally,we combine the hyperspectral dictionary with the sparse coefficients of multispectral to obtain the hyperspectral image with high spatial resolution.(2)In order to make full use of the spatial structure information and spectral information of multi-spectral images,a hyperspectral and multi-spectral image fusion method based on global structure and spectral self-similarity is proposed.The algorithm adopts a coupled sparse representation strategy,which minimizes the reconstruction errors of hyperspectral images and multispectral images at the same time.Compared with the method of using multispectral images to solve the sparse representation problem to obtain sparse coefficients,the reconstruction error is smaller,and the reconstruction space is high.The resolution hyperspectral image is closer to the reference image.In the experimental part of the thesis,the proposed algorithm is compared with the related algorithms in recent years on multiple public data sets,and the effectiveness of the proposed algorithm is verified through qualitative and quantitative analysis.