Analyzing the Contribution of Training Algorithms on Deep Neural Networks for Hyperspectral Image Classification

Abstract

Autoencoder (<small>AE</small>)-based deep neural networks learn complex problems by generating feature-space conjugates of input data. The learning success of an AE is too sensitive for a training algorithm. The problem of hyperspectral image (<small>HSI</small>) classification by using spectral features of pixels is a highly complex problem due to its multi-dimensional and excessive data nature. In this paper, the contribution of three gradient-based training algorithms (i.e., scaled conjugate gradient (<small>SCG</small>), gradient descent (<small>GD</small>), and resilient backpropagation algorithms (<small>RP</small>)) on the solution of the HSI classification problem by using AE was analyzed. Also, it was investigated how neighborhood component analysis affects classification performance for training algorithms on HSIs. Two hyperspectral image classification benchmark data sets were used in the experimental analysis. Wilcoxon signed-rank test indicates that RB is statistically better than SCG and GD in solving the related image classification problem.