Hyperspectral Image Classification Method Based on Neighbor Propagation Clustering and Sparse Multi-kernel Learning

Abstract

The invention discloses a hyperspectral classification method based on neighbor propagation clustering and sparse multi-kernel learning, which mainly solves the problem of poor hyperspectral image classification performance in the prior art. The implementation plan is as follows: firstly, use the training samples in all bands to construct a set of kernel matrices; secondly, use the nearest neighbor propagation method for clustering, and select a subset of kernel matrices with high discrimination and low redundancy; thirdly, use the selected kernel matrix The subset, through the multi-kernel learning method with sparse constraints, learns the kernel weights and support vector coefficients; finally, uses the learned classifier to classify the unknown hyperspectral image. The multi-kernel learning classification method adopted in the present invention can process complex hyperspectral data with variable local distribution by using multiple kernels with different functions and different parameters, and obtain high-precision hyperspectral image classification results, which can be used for agricultural monitoring, geological exploration, Distinguishing and identification of ground features in disaster environment assessment and other fields.

Description

technical field [0001] The invention belongs to the technical field of remote sensing image processing, and in particular relates to a hyperspectral image classification method, which can be used to distinguish and distinguish ground objects in the fields of agricultural monitoring, geological exploration, disaster environment assessment and the like. Background technique [0002] Over the past thirty years, with the development of science and technology, remote sensing technology has also been greatly developed. The hyperspectral remote sensing system occupies an extremely important position in the field of earth observation. Hyperspectral remote sensing technology is a new type of remote sensing technology developed on the basis of multispectral remote sensing technology. Compared with multispectral images, hyperspectral images can provide richer spectral information of ground features. Hyperspectral images can obtain approximately continuous spectral information of targ...

AI Technical Summary

Problems solved by technology

However, due to the huge kernel size, the multi-kernel learning method has high computational complexity, and it is difficult to efficiently process complex hyperspectral image data.

In addition, in hyperspectral image data, when the number of labeled samples is limited, a large number of spectral bands will cause the Hughes phenomenon, resulting in the degradation of the classification performance of multi-kernel learning.

[0006] In summary, the existing multi-kernel learning classification methods are directly used for hyperspectral image classification, but there are problems of excessive kernel size and poor classification accuracy.

Images