Hyperspectral image classification method based on convolutional neural network and spatial spectrum information fusion

Abstract

The invention discloses a hyperspectral image classification method based on a convolutional neural network and spatial spectrum information fusion. The method comprises the steps that S1 X and Y axiscoordinates of each pixel point of a hyperspectral image are extracted as spatial information, and the spatial information and spectral information are combined as the feature information of a sample; S2 training set and test set data are randomly divided; S3 training set samples including the spatial information and the spectral information are put into a one-dimensional convolutional neural network to train a classification model; and S4 the training set samples including the spatial information and the spectral information are put into the classification model for classification prediction. The convolutional neural network uses convolution kernels of different sizes to carry out convolution operation, which can effectively extract feature information with different resolutions in the spectral dimension of a hyperspectrum. In addition, the spectral dimension information and the spatial dimension information are input into the neural network to learn simultaneously. The feature of dual high resolutions of the hyperspectrum is fully used. The algorithm structure is simple, and the classification accuracy can be significantly improved.

Description

technical field [0001] The invention relates to the technical field of hyperspectral image processing, in particular to a hyperspectral image classification method based on fusion of convolutional neural network and spatial spectrum information. Background technique [0002] Hyperspectral remote sensing data combines spectral technology and imaging technology to obtain a three-dimensional data block that combines two-dimensional spatial data and one-dimensional spectral data, and has the "double high characteristics" of high spectral resolution and high spatial resolution. Hyperspectral data has broad application prospects for space target detection, soil composition analysis, and vegetation type identification. Therefore, the hyperspectral image classification problem has been receiving much attention. Traditional hyperspectral image classification generally uses spectral dimension information to classify pixels. Commonly used methods include: Support Vector Machine (SVM),...

AI Technical Summary

Problems solved by technology

This method makes good use of the information of the spectral dimension and the spatial dimension, and the classification accuracy is very high. However, there are still problems such as large amount of training, long time consumption, and difficult training when the amount of data is small.

Images