A Comparison Between Support Vector Machine (SVM) and Convolutional Neural Network (CNN) Models For Hyperspectral Image Classification

Abstract

This study presents a methodology model for the spectral classification of hyperspectral images. The applied methodology, first extracts neighbouring spatial regions via a suitable statistical support vector machine (SVM-Linear) architecture, support vector machine radial basis function (SVM-RBF) and Deep Learning (DL) architecture that comprises principal component analysis (PCA) and convolutional neural networks (CNN) and then applies a soft max classifier. PCA is introduced to reduce the high spectral dimensionality, noise, and redundant information of the input image. The SVM-Linear, SVM-RBF and CNN model is used to extract useful high-level features automatically given that it provides results comparable with each other, including hyperspectral image classification. However, because the CNN, SVM models alone may fail to extract features with different scales and to tolerate the large-scale variance of image objects, the presented methodology uses PCA optimization for spatial regions to construct features that can be then used by the SVM and CNN model to classify hyperspectral images. Experimental results obtained in the classification of the Hyperspec-VNIR Chikusei datasets show that the performance of the presented model is competitive with that of other DL and traditional machine-learning methods. The best overall accuracy of the presented methodology for the Hyperspec-VNIR Chikusei datasets is 98.84 % in SVM-RBF model.