Hyperspectral-remote-sensing-image classification technology combining spectral, spatial and hierarchical information

Abstract

The invention discloses a hyperspectral-remote-sensing-image classification technology combining spectral, spatial and hierarchical information. The technology includes: inputting a to-be-classified hyperspectral image, and extracting a reference data sample set; selecting a training sample set of supervised classification; using principal component analysis to extract top-three principal component wavebands of the image, and using an extension morphology method to extract spatial feature vectors; using Markov feature selection to carry out dimension reducing on the original image, using the algebraic multi-grid method to construct the image pyramid, and using a hierarchical segmentation method to obtain multi-layer segmentation results; combining original spectral feature vectors, the spatial feature vectors and hierarchical feature vectors to construct multiple kernel matrices; using a support vector machine method to calculate kernel matrix differences to obtain final class attribute labels; and outputting a final image classification graph. The invention provides an effective classification method, which can fully extract and mine image information, and can effectively improveclassification accuracy of the hyperspectral image.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a hyperspectral image classification method combining spectral, spatial and hierarchical information. Background technique [0002] Compared with multispectral remote sensing images, hyperspectral remote sensing images have richer information, which can accurately reflect the attribute differences between different types of ground objects, realize accurate extraction and identification of ground objects, and provide more accurate hyperspectral remote sensing image analysis. Lay a good foundation with industry applications. However, hyperspectral images have high dimensionality, high band correlation, noise and mixed pixels, and limited ground truth data references, which bring great challenges to the analysis and processing of hyperspectral remote sensing information. The single-kernel image classification method based on spectral features only determines the classifica...

AI Technical Summary

Problems solved by technology

However, hyperspectral images have high dimensionality, high band correlation, noise and mixed pixels, and limited ground truth data references, which bring great challenges to the analysis and processing of hyperspectral remote sensing information.

The single-kernel image classification method based on spectral features only determines the category of pixels based on the spectral features of ground objects, and does not use the spatial information of the image, so the classification accuracy obtained by this type of method is difficult to further improve

The composite nuclear image classification method based on spectral and spatial information is based on the combination of hyperspectral image spatial information (mainly including texture information, spatial structure information, object size information, object outline information, spatial distribution information, etc.) and object spectral features. In general, more accurate classification results can be obtained, but often only one extraction method is used for spatial features, and it is difficult to fully obtain all structural information

[0003] Traditional compound kernel methods are often not enough to extract spatial information. To solve this problem, multi-scale correlation methods have attracted researchers' attention in recent years.

Images