Hyperspectral classification method combining graph structure and convolutional neural network

Abstract

The invention relates to a hyperspectral image classification method combining a graph structure and a convolutional neural network. The convolutional neural network can decompose three dimensions into one-dimensional and two-dimensional convolutional neural networks, and the superpixel graph network can be used for subsequent multi-scale two-dimensional processing; the method comprises five steps of hyperspectral image segmentation and spectral feature extraction, multi-scale feature extraction based on a graph network, multi-scale feature extraction based on a convolutional neural network, feature fusion and pixel classification, and loss function and model training. Compared with the prior art, the invention provides a novel multi-scale fusion network and a spectral transformation mechanism, the graphic features and local pixel features based on multi-scale superpixels can be extracted, and the spectral features of graph nodes are extracted in a one-dimensional manner; the method can restrain the noise of an original hyperspectral image, improve the adaptability of the image convolutional neural network to different hyperspectral images, improve the classification precision, automatically extracts the hyperspectral features, and complete the classification. And the classification accuracy reaches over 93%.

Description

technical field [0001] The invention relates to the technical field of geographic remote sensing, and relates to a hyperspectral image classification method combining a graph structure and a convolutional neural network. Background technique [0002] Hyperspectral images collected from satellites or airborne contain hundreds of contiguous bands, containing rich spectro-spatial information. Due to the unique advantage of hyperspectral, which can distinguish land cover classes at the pixel level, hyperspectral classification that classifies each pixel as a specific label has attracted much attention. Hyperspectral classification has been applied in various fields such as land management, environmental monitoring, military survey, and agricultural assessment. However, hyperspectral high dimensionality, insufficient labeled training samples, and complex spectral noise effects bring great difficulties to hyperspectral classification, and various methods have been studied to solv...

AI Technical Summary

Problems solved by technology

[0006] The basic idea of ​​the present invention is: firstly, a hyperspectral image multi-feature fusion network combining graph network and convolutional neural network is proposed. Extracting multi-scale spatial features from hyperspectral images not only solves the problem of superpixel HSI feature extraction, but also integrates the two networks of graph network and convolutional neural network, and then proposes a dual-core convolutional neural network for extracting from superpixel HSI. Extract multi-scale local features of pixels, and finally, introduce cascade operation to fuse multi-scale features of four branches

Images