Research On Sparse-Representation Based Hyperspectral Image Classification Algorithms

Hyperspectral images have the characteristics of high spatial resolution,wide spectral band coverage,and accurate feature recognition capabilities.In recent years,with the rapid development of hyperspectral imaging technology and hyperspectral classification theory,the spatial spectrum information of hyperspectral images has been fully mined,and it is widely used in precision agriculture,geological prospecting,national defense technology,environmental monitoring,etc.field.However,in the application process,there are still some problems to be solved,including high-dimensional redundancy of data and high band correlation.In recent years,classification algorithms based on sparse representation have developed rapidly in the field of hyperspectral image classification.Aiming at the problems of existing sparse representation methods such as dictionary redundancy and insufficient utilization of multi-feature information,this paper proposes a hyperspectral image classification algorithm based on multi-feature LC-KSVD;in view of the defect of sparse classification model spectral vector Euclidean distance loss,A hyperspectral image classification algorithm based on local information constraints and sparse representation is proposed.The main research contents of this article are:Multi-feature baesd LC-KSVD algorithm for hyperspectral image classification is proposed.Aiming at the problem of complex hyperspectral image background and large-scale dictionary,the algorithm introduces complementary multi-feature information into the sparse model,and uses the K-SVD algorithm to train a more concise and robust multi-feature dictionary.In addition,in the model training stage,this paper introduces a multi-feature classifier,which can automatically determine the relative importance of different features for classification,and achieve the optimal combination of features,which is conducive to more accurate feature classification.Finally,combine the multi-feature information of the pixel to be measured and the multi-feature classifier to classify.Experiments show that the hyperspectral image classification algorithm based on multi-feature LC-KSVD has higher classification accuracy.Locality-constrained sparse representation for hyperspectral image classification is proposed.This algorithm improves on the existing sparse representation algorithms that cannot solve the problems of "same spectra with different spectra" and "same spectra with different spectra" in hyperspectral images.The algorithm introduces a variety of distance information to distinguish the difference of the ground feature categories.First,the nearest neighbor algorithm（KNN）is applied to the training data set,and the local constraint dictionary is formed by excluding the samples of the separated test pixels in the Euclidean space.Then,use the category label information to perform Orthogonal Match Pursuit（OMP）algorithm to obtain the corresponding sparse coding.Finally,the classification is carried out by the principle of least error.Experiments are performed on three commonly used hyperspectral classification data sets.The hyperspectral image classification algorithm based on local information constraints and sparse representation can effectively improve the classification efficiency.