Hyperspectral classification method and system based on graph structure

Abstract

The invention relates to a hyperspectral classification method and system based on a graph structure. The method comprises the following steps: segmenting a hyperspectral image into N superpixels; each super pixel comprises a plurality of pixels; constructing an adjacent matrix of the graph according to the N superpixels; each element in the adjacent matrix represents the relationship between the features of each superpixel; according to the adjacent matrix, performing feature extraction on the hyperspectral image by using a double-layer graph convolution algorithm to obtain a first feature of each superpixel; learning a first feature of each superpixel by using a self-attention mechanism to obtain a second feature of each superpixel; and classifying each superpixel in the hyperspectral image according to each second feature. According to the invention, the accuracy of hyperspectral classification is improved.

Description

technical field [0001] The invention relates to the field of spectral classification, in particular to a hyperspectral classification method and system based on a graph structure. Background technique [0002] Hyperspectral images have a large number of spectral bands, contain rich spatial and spectral information, and can accurately identify objects containing different materials. Compared with multispectral imagery or RGB (red, green, and blue) analysis, hyperspectral image analysis can identify object features more effectively. Therefore, hyperspectral image classification, which classifies each image pixel into a specific label, has attracted great attention in many fields, such as agricultural monitoring, military reconnaissance, and disaster prevention and control. However, problems such as multi-band hyperspectral images, spatial variability of spectral features, and difficulty in obtaining labels have brought great difficulties to hyperspectral classification. [0...

AI Technical Summary

Problems solved by technology

However, the convolutional neural network method requires a large number of training labels, and the hyperspectral image label data is small, and it is difficult to provide a large number of training samples.

In addition, convolutional neural network kernels are mainly designed for regular pattern recognition, so they cannot adaptively capture irregular geometric changes in different object regions in hyperspectral images

Finally, the weight of the convolution kernel of the convolutional neural network is fixed, which causes the network to cause edge loss during the feature extraction process, and may cause classification errors during the classification process.

[0005] Therefore, both traditional machine learning methods and convolutional neural networks in deep learning face certain limitations in hyperspectral classification.

Images