Research On Spatial-Spectrum Hyperspectral Image Classification Algorithm Based On SPCA

Hyperspectral image classification is to assign a category label to each pixel in the data volume,so as to determine the classification of ground objects And its distribution.The fast and highly accurate hyperspectral remote sensing image classification algorithm is Prerequisites for realizing various practical applications.Hyperspectral image classification has always received great attention from scientific and technical personnel of remote sensing information processing.A large number of researchers are constantly exploring new methods and improving the original methods in order to continuously improve the accuracy and speed of hyperspectral image classification algorithms.Among them,spatial-spectrum classification has been paid more and more attention.Segmented principal component analysis can achieve the effects of reducing the number of bands and reducing the amount of calculations while retaining the spatial structure.After the segmented principal components are reduced in dimension,the hyperspectral image can be effectively extracted by using two spatial feature extraction methods: domain transform recursive filtering and local binary mode,achieving the purpose of combining spatial and spectral information.Based on this,this paper mainly studies two hyperspectral image classification algorithms based on segmented principal component analysis and spatial feature extraction:1.A hyperspectral image classification algorithm based on segmented principal component analysis and domain transform recursive filtering is given.A new hyperspectral image classification algorithm based on segmented principal component analysis(SPCA)and domain transform recursive filtering(DTRF)is proposed.First,the SPCA method is used to reduce the dimensionality of the hyperspectral image and extract the first principal component of each band subset.Then,a domain transformation recursive filter with different parameters is used to filter the first principal component of each band subset to form a stacked edge-preserving filter map,and the function of spatial feature extraction is realized.Principal component analysis(PCA)is used to fuse the edge-preserved filter maps for feature fusion to reduce redundant information and enhance the separation between classes.Finally,the basic threshold classifier(BTC)is used to classify the fused principal components.Simulation experiments show that the proposed method can improve classification accuracy,and is superior to existing methods in terms of overall classification accuracy,average classification accuracy,and Kappa coefficient.2.A hyperspectral image classification algorithm based on SPCA and Local Binary Pattern(LBP)is designed.By combining segmented principal component analysis and local binary mode for feature extraction of hyperspectral images,it ensures the effective removal of interspectral redundancy in hyperspectral images,while protecting the spatial local neighborhood information of hyperspectral images.PCA fusion is then used to reduce redundancy while enhancing separation between classes.For this reason,this algorithm not only can fully exploit the inter-spectral spatial characteristics of hyperspectral images,improve the classification accuracy and Kappa coefficient to a large extent,but also has good classification performance in the case of small samples.Simulation experiments show that the proposed algorithm has certain advantages over existing classical hyperspectral image classification algorithms in terms of objective evaluation.