Research On Hyperspectral Image Classification Method Based On Sparse Representation

Hyperspectral image data effectively combines the image information reflecting the spatial geometric relationship of ground objects with the spectral information reflecting the radiation properties of ground objects.Hyperspectral images usually consist of hundreds of adjacent and narrow bands.Hyperspectral images are usually composed of hundreds of adjacent and narrow bands,from the visible spectrum of the same scene to the near-infrared spectrum,which provide rich land cover information and have attracted extensive attention in land use analysis,urban mapping,military reconnaissance,and environmental monitoring.In these applications,the most basic problem is classification.Based on the known hyperspectral image classification of joint sparse representations,this thesis further studies the dictionary of sparse representations,and proposes a classification of hyperspectral image classification based on joint sparse representations of the secondary dictionary and a joint sparse representation based on dictionary optimization.Hyperspectral image classification.The main research contents of this thesis are as follows:1.To solve the problem of insufficient utilization of spectral information by the joint sparse representation hyperspectral image classification algorithm,the algorithm named a secondary dictionary joint sparse representation classification algorithm is proposed.The dictionary is usually made of the pixels directly in most of the sparse representation hyperspectral image classification algorithm,without further processing.In order to improve the probability that the atom of true class is chosen to reconstruct the test pixel,the thesis updates dictionary by using the formula which is changed from the formula of gravity.Multiply atom and the gravitational value which is calculated by atom and test pixel to increase the probability of correct classification.In order to make full use of the spectral information,the exponential smoothing formula is used to deal with the residual.Experiments on Indian Pines and Salinas data sets show that the proposed algorithm is effective in improving classification accuracy.2.To solve the problem of complex spectral information of hyperspectral images,large amount of sparse representation dictionary data and save the cost of information collection of hyperspectral data sets,a joint sparse representation based on dictionary optimization was proposed.It is based on kernel joint sparse representation classification algorithm.Initially select a small number of atoms,calculate the spectral similarity between each atom and the cluster center of the sample through the Gaussian kernel function,and take the average value.Increase the number of atoms for the atomic dictionary with low spectral similarity to make it sufficiently representative of the class atom.The PCA transform is used to extract the principal components of the dictionary after reselecting the atoms,reducing the redundant components of the dictionary,which is convenient for sparse representation classification.Experiments on the Indian Pines and Salinas data sets show that the method can better classify hyperspectral data sets with fewer atoms.