Method for extracting combined kernel minimum noise fraction characteristic of high spectral image

Abstract

The invention provides a method for extracting a combined kernel minimum noise fraction characteristic of a high spectral image, wherein the method belongs to the technical field of high spectral image data processing and application. The method comprises the following steps of 1), acquiring high spectral reflectivity data; 2), estimating image noise; 3) constructing a least noise separation transform model; 4) constructing a dual mode least noise separation transform model; 4), constructing a combined kernel function through using advantages of high learning capability of a Gaussian kernel function and high generalization capability of a polynomial kernel function; 5), constructing a combined kernel minimum noise fraction model; and 6), utilizing combined kernel minimum noise fraction forextracting the characteristic of the high spectral image. According to the method of the invention, through the combined kernel function, the original non-separatable high spectral data are mapped toa kernel characteristic space so that the high spectral data are separatable, thereby obtaining a high spectral image characteristic extraction effect which is better than kernel minimum noise separation transform and traditional least noise separation transform.

Description

technical field [0001] The invention relates to the technical field of hyperspectral image data processing and application, in particular to a hyperspectral image combination kernel minimum noise separation transformation feature extraction method. Background technique [0002] Hyperspectral remote sensing combines imaging technology and spectral technology, and is a frontier research field of remote sensing. Hyperspectral images have spectral resolution up to nanometer scale and have a wide range of applications. However, it has problems such as large number of bands, high correlation between bands, nonlinear data structure, and curse of dimensionality. [0003] Hyperspectral image feature extraction is a key link in hyperspectral image data processing. It converts samples in high-dimensional space to low-dimensional space through mapping or transformation to achieve dimensionality reduction and data redundancy. It can be divided into There are two types of linear feature...

AI Technical Summary

Problems solved by technology

Since the kernel method was successfully applied in support vector machines in the mid-1990s, many scholars have proposed nonlinear feature extraction methods based on kernel methods, such as kernel principal component analysis, kernel Fisher discriminant analysis, etc.; it realizes nonlinear feature extraction through kernel functions. For linear mapping, different kernel functions have different advantages and disadvantages. For example, the widely used Gaussian kernel function is a typical local kernel function. Although it has strong learning ability, its generalization ability is weak; the polynomial kernel function is a global kernel function. , weak learning ability, but strong generalization ability

Using a kernel function alone often cannot get a good hyperspectral image feature extraction effect

Images