A multi-scale spatial-spectral collaborative classification method for hyperspectral images

Abstract

A multi-scale space-spectrum cooperative classification method for hyperspectral images relates to the field of hyperspectral image information technology processing. The method solves the problem that the single method of classifying hyperspectral images using spectral information or spatial information in the prior art is easy to cause loss of target information, resulting in low classification accuracy. Step 1: Perform feature extraction on the original hyperspectral image H to obtain a spectral information set H composed of band subsets spec ; Step 2: For spectral information set H spec Extract multi-scale spatial information to obtain multiple sets of multi-scale spatial information data sets H spet ; Step 3: Combine multiple sets of multi-scale spatial information datasets H spet with spectral information set H spec Perform fusion and preliminary classification, and obtain the preliminary classification result Figure Q init ; Step 4: Map the preliminary classification results to Q init Decompose into k probability result maps P; Step 5: Post-process the k probability result maps P to obtain the final classification result map O fin , so as to complete the spatial spectrum collaborative classification of hyperspectral images. The present invention is mainly used for space spectrum classification of hyperspectral images.

Description

technical field [0001] The invention relates to the technical processing field of hyperspectral image information. Background technique [0002] Hyperspectral images have the characteristics of "space-spectrum integration" and contain rich spatial and spectral information. At present, many commonly used classifiers only use spectral information for classification, and lack of consideration for spatial information. However, ground targets in images often have Certain spatial correlation and continuity; [0003] The spatial information used in most existing research methods or experiments is often single-scale, that is, the spatial information is usually extracted by a single fixed-window spatial filter. Since the sizes of various ground targets are not the same, only Using a single-scale filter to extract spatial information is likely to cause the loss of ground targets, especially small-scale target information, and these small-scale targets are often quite important. [0...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to solve the problem in the prior art that only using spectral information or spatial information to classify hyperspectral images can easily cause loss of target information and lead to low classification accuracy. The present invention provides a hyperspectral image classification method. Multi-scale spatial-spectral collaborative classification method