Hyperspectral image classification method of graph convolution network based on multi-graph structure

Abstract

The invention discloses a hyperspectral image classification method of a graph convolution network based on a multi-graph structure. The hyperspectral image classification method comprises the following steps: reading a hyperspectral image cube and a training set; rearranging the hyperspectral data to obtain a pixel data matrix; constructing a forced nearest neighbor connection matrix and a spatial nearest neighbor connection matrix according to the pixel data matrix; performing convolution on the forced nearest neighbor connection matrix, the spatial nearest neighbor weight matrix and the pixel data matrix through a graph convolution network, and obtaining a feature matrix; and splicing the feature matrixes, classifying the feature matrixes AF of the pixels by using a softmax classifier,and finally verifying a hyperspectral image classification result.

Description

technical field [0001] The invention relates to the technical field of high-dimensional image processing, in particular to a hyperspectral image classification method based on a graph convolutional network with a multi-graph structure. Background technique [0002] A hyperspectral image is an image cube that is acquired by a hyperspectral image sensor and has a large number of spectral bands. Each pixel is composed of spectral features within a certain range of spectral bands. The spectral feature is an essential feature of the target ground object, which reflects the characteristics of the material composition of the ground object. Compared with traditional single-band images and multispectral images, hyperspectral images have hundreds of spectral bands and contain rich spectral information. At the same time, the large amount of data and high-dimensional features of hyperspectral images also bring great challenges to image processing and classification tasks. In addition,...

AI Technical Summary

Problems solved by technology

[0003] In the existing graph convolutional network classification method for hyperspectral images, good classification accuracy can be obtained, but when the number of labeled samples is small, the classification accuracy is not ideal

The hyperspectral image classification method based on the graph convolutional network has certain requirements on the structure of the graph and the number of labeled samples. In the case of few training samples, how to make full use of the spectral information and spatial information of the hyperspectral image to design the graph. structure to improve classification accuracy is a challenge

Images