A hyperspectral image classification algorithm based on 3D convolution spectral space feature fusion

Abstract

The invention discloses a hyperspectral image classification algorithm based on 3D convolution spectral space feature fusion, and the algorithm reduces the correlation between spectral features through employing PCA whitening, and greatly improves the training speed of a neural network. According to the method, the data is subjected to feature extraction in a manner of combining 3D convolution operation and 1D convolution operation; compared with the prior art, the method has the advantages that the classification precision can be higher, the representation effect on small samples is better, the calculation efficiency is higher, no information loss exists in the feature fusion process, the efficient and accurate classification can be completed, and in addition, the classification precisionis obviously promoted by using the 3D convolution hyperspectral image spectral space feature fusion method.

Description

technical field [0001] The invention relates to the technical field of hyperspectral image classification, in particular to a hyperspectral image classification algorithm based on 3D convolution spectral-space feature fusion. Background technique [0002] In the field of hyperspectral image classification, support vector machine (SVM) is considered the most popular method due to its high efficiency. It is a supervised learning model that maps the sample space to a high-dimensional In the feature space, the nonlinear separable problem in the original sample space is transformed into a linear separable problem in the feature space, so as to realize the classification of hyperspectral images. However, this method has low classification accuracy and cannot effectively deal with image The existing noise and the accuracy in the small sample environment cannot meet the actual application requirements; therefore, in recent years, a hyperspectral image spectral-space feature fusion m...

AI Technical Summary

Problems solved by technology

[0002] In the field of hyperspectral image classification, support vector machine (SVM) is considered the most popular method due to its high efficiency. It is a supervised learning model that maps the sample space to a high-dimensional In the feature space, the nonlinear separable problem in the original sample space is transformed into a linear separable problem in the feature space, so as to realize the classification of hyperspectral images. However, this method has low classification accuracy and cannot effectively deal with image The existing noise and the accuracy in the small sample environment cannot meet the actual application requirements; therefore, in recent years, a hyperspectral image spectral-space feature fusion method using the principal component analysis method to reduce the dimensionality and using Gabor features has emerged to improve the classification accuracy , although this method can improve the classification accuracy of hyperspectral images to a certain extent, it often sacrifices part of the information carried in the data, making the information after spectral-space fusion incomplete, which limits the upper limit of its performance.

Images