Citation: Tinega, H.C.; Chen, E.; Ma,
L.; Nyasaka, D.O.; Mariita, R.M.
HybridGBN-SR: A Deep 3D/2D
Genome Graph-Based Network for
Hyperspectral Image Classification.
Remote Sens. 2022, 14, 1332.
https://doi.org/10.3390/rs14061332
Academic Editors: Pedro
Latorre-Carmona and Antonio
J. Plaza
Received: 22 February 2022
Accepted: 7 March 2022
Published: 9 March 2022
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2022 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
remote sensing
Article
HybridGBN-SR: A Deep 3D/2D Genome Graph-Based Network
for Hyperspectral Image Classification
Haron C. Tinega
1
, Enqing Chen
1,
*, Long Ma
1
, Divinah O. Nyasaka
2
and Richard M. Mariita
3
1
School of Information Engineering, Zhengzhou University, Zhengzhou 450001, China;
tinegaharon@gmail.com (H.C.T.); ielongma@zzu.edu.cn (L.M.)
2
The Kenya Forest Service, Nairobi P.O. Box 30513-00100, Kenya; dondieki@kenyaforestservice.org
3
Microbial BioSolutions, Troy, New York, NY 12180, USA; richard.mariita@microbialbiosolutions.com
* Correspondence: ieeqchen@zzu.edu.cn; Tel.: +86-371-6778-1544
Abstract: The successful application of deep learning approaches in remote sensing image classifica-
tion requires large hyperspectral image (HSI) datasets to learn discriminative spectral–spatial features
simultaneously. To date, the HSI datasets available for image classification are relatively small to
train deep learning methods. This study proposes a deep 3D/2D genome graph-based network
(abbreviated as HybridGBN-SR) that is computationally efficient and not prone to overfitting even
with extremely few training sample data. At the feature extraction level, the HybridGBN-SR utilizes
the three-dimensional (3D) and two-dimensional (2D) Genoblocks trained using very few samples
while improving HSI classification accuracy. The design of a Genoblock is based on a biological
genome graph. From the experimental results, the study shows that our model achieves better
classification accuracy than the compared state-of-the-art methods over the three publicly available
HSI benchmarking datasets such as the Indian Pines (IP), the University of Pavia (UP), and the
Salinas Scene (SA). For instance, using only 5% labeled data for training in IP, and 1% in UP and SA,
the overall classification accuracy of the proposed HybridGBN-SR is 97.42%, 97.85%, and 99.34%,
respectively, which is better than the compared state-of-the-art methods.
Keywords: convolutional neural networks; genome graph; hyperspectral image classification; remote
sensing; remote sensing image classification; residual learning; spectral–spatial features
1. Introduction
Remote sensing works by moving a vision system (satellite or aircraft) across the
Earth’s surface at various spatial resolutions and in different spectral bands of the magnetic
spectrum to capture hyperspectral images (HSI) [1]. The vision system uses both imaging
and spectroscopic methods to spatially locate specific components within the image scene
under investigation based on their spectral features. The collected HSI data are a three-
dimensional data structure with the x and y axes capturing the dimensions of the spatial
images, and the z-axis is the number of spectral bands. Consequently, each pixel located on
the x–y spatial domain contains a label representing the physical land cover of the target
location [2].
For feature extraction and classification purposes, the voluminous spectral–spatial
cues present in the HSI image represent an advantage in the detailed representation of
the analyzed samples. However, they contain high spectral redundancy caused by sig-
nificant interclass similarity and intraclass variability caused by changes in atmospheric,
illumination, temporal, and environmental conditions, leading to data handling, stor-
age, and analysis challenges [2]. For instance, an HSI system with a spatial resolution
of 145 × 145 pixels will produce an image with 21,025 pixels for one spectral band. If
the data contain 200 spectral bands, then a single image would produce over 4 million
(145 × 145 × 200) data points. To overcome the challenges of spectral redundancy, most of
Remote Sens. 2022, 14, 1332. https://doi.org/10.3390/rs14061332 https://www.mdpi.com/journal/remotesensing