High-spectral image classification method

Abstract

The invention discloses a high-spectral image classification method, and relates to the technical field of information processing. An image to be classified is segmented in multiple dimensions, a salient map is extracted from the image after multi-dimension segmentation, features are extracted from the image from which the salient map is extracted, the extracted image features are normalized, and the normalized image features are classified in a nonlinear kernel function method; and the nonlinear kernel function method concretely refers to one selected from a Gaussian process, a support vector machine, kernel principal component analysis, a kernel function Fisher discriminant method and a kernel projection pursuit method. The image data bulk is reduced greatly, and the classification speed is high; and via nonlinear mapping, influence of radiation and geology errors is reduced greatly, the high-spectral image classification precision is higher, and the method of the invention is suitable for application occasions of the high-spectral remote-sensing images.

Description

technical field [0001] The present invention relates to the technical field of information processing, and in particular to a hyperspectral image classification method. Background technique [0002] After the development of remote sensing technology in the latter half of the 20th century, major changes have taken place in theory, technology and application. Remote sensing image classification is one of the key technologies in remote sensing geographic information system. Fast, high-precision remote sensing image automatic classification algorithm is the key to realize the dynamic monitoring, evaluation and forecasting of the environment. Hyperspectral remote sensing images are spectral remote sensing images with a spectral resolution of 10-20 nm, which can obtain spectral information of hundreds of surface object bands. They have many bands, nonlinearity, coexistence of spatial correlation and inter-spectral correlation, and are difficult to obtain. Sample marking and other...

AI Technical Summary

Problems solved by technology

Radiation errors can be compensated by certain calculation methods, but geometric errors are difficult to completely remove because of their complex factors, which makes hyperspectral images have nonlinear characteristics to varying degrees, making it difficult to perform linear simulation of images combined, it is also difficult to correctly classify hyperspectral images with linear classifiers

In addition, hyperspectral data has a lot of redundancy in space and band

Images