Unsupervised hyperspectral data dimension reduction method based on block low-rank tensor model

Abstract

The present invention provides an unsupervised hyperspectral data dimension reduction method based on a block low-rank tensor model and mainly solves the problems of difficulty in acquiring hyperspectral data type labels and insufficient utilization for hyperspectral image spectral-spatial cooperative information in the prior art. The present invention adopts the technical scheme that 1, hyperspectral image data is input and a tensor expression form of a hyperspectral image is acquired; 2, a spatial block size of the hyperspectral image is set and the hyperspectral image is subjected to spatial mirror extension; 3, the extended hyperspectral image is subjected to block processing in the spatial dimension; 4, a three-order image block is subjected to the clustering operation; 5, each cluster group is respectively subjected to dimension reduction by low-rank tensor analysis to obtain a dimension reduction result of each cluster group; and 6, the three-order image block subjected to dimension reduction is reduced according to an original position so as to obtain an original hyperspectral image dimension reduction result. According to the unsupervised hyperspectral data dimension reduction method based on the block low-rank tensor model, separability of data subjected to dimension reduction is improved and the unsupervised hyperspectral data dimension reduction method is beneficial for improving accuracy of subsequent hyperspectral image classification and can be used for processing the hyperspectral image.

Description

technical field [0001] The invention belongs to the field of image processing, and further relates to a data dimensionality reduction method, which can be used for dimensionality reduction and classification of remote sensing image data. technical background [0002] Hyperspectral images contain a wealth of spatial and spectral information, which characterizes the image features of the spatial distribution of ground objects, the radiation intensity and spectral characteristics of pixels, and the space-spectral structure of hyperspectral images such as figure 1 shown. Hyperspectral images are rich in information and have broad application prospects in many fields, such as surveying and mapping, geology, agriculture, forestry, resource and environment monitoring and management, military reconnaissance and identification of camouflage and other fields. In recent years, with the development of technology, hyperspectral image resources have become increasingly abundant, and the ...

AI Technical Summary

Problems solved by technology

This method requires that the hyperspectral image must have a strong global spatial correlation. If the global spatial correlation of the image is not obvious, or if there is only a strong local spatial correlation, a good dimensionality reduction effect cannot be obtained.

[0006] The patent "A method for dimensionality reduction and classification of hyperspectral images based on block low-rank tensor analysis" (application number: 201210403361.1, application publication number: CN102938072A, publication date: 2010.10.20) disclosed a method based on A low-rank tensor hyperspectral image dimensionality reduction method for blocks, which introduces the block idea into a hyperspectral image dimensionality reduction method based on low-rank tensor analysis, transforms the global spatial correlation constraints into local spatial correlation constraints, To a certain extent, it overcomes the negative impact of the weak global spatial correlation of the image on the dimensionality reduction effect, but this invention does not take into account that hyperspectral images usually have strong non-local correlations in addition to local correlations. At the same time, the invention When the invention uses a supervised classifier to classify the data after dimension reduction, it is necessary to set the threshold of the number of samples, which increases the complexity of subsequent applications

Images