Towards reduced-cost hyperspectral and multispectral image classification

Abstract

In recent years, Hyperspectral Imaging systems (HSI) have become a powerful source for reliable data in applications such as remote sensing, agriculture, and biomedicine. However, the abundant spectral and spatial information of hyperspectral images makes them highly complex, which leads to the need for specialized Machine Learning algorithms to process and classify them. In that sense, the contribution of this thesis is multi-folded. We present a low-cost convolutional neural network designed for hyperspectral image classification called Hyper3DNet. Its architecture consists of two parts: a series of densely connected 3-D convolutions used as a feature extractor, and a series of 2-D separable convolutions used as a spatial encoder. We show that this design involves fewer trainable parameters compared to other approaches, yet without detriment to its performance. Furthermore, having observed that hyperspectral images benefit from methods to reduce the number of spectral bands while retaining the most useful information for a specific application, we present two novel hyperspectral dimensionality reduction techniques. First, we propose a filter-based method called Inter-Band Redundancy Analysis (IBRA) based on a collinearity analysis between a band and its neighbors. This analysis helps to remove redundant bands and dramatically reduces the search space. Second, we apply a wrapper-based approach called Greedy Spectral Selection (GSS) to the results of IBRA to select bands based on their information entropy values and train a compact Convolutional Neural Network to evaluate the performance of the current selection. We also propose a feature extraction framework that consists of two main steps: first, it reduces the total number of bands using IBRA; then, it can use any feature extraction method to obtain the desired number of feature channels. Finally, we use the original hyperspectral data cube to simulate the process of using actual filters in a multispectral imager. Experimental results show that our proposed Hyper3DNet architecture in conjunction with our dimensionality reduction techniques yields better classification results than the compared methods, producing more suitable results for a multispectral sensor design.