A Comparative Review of Deep Learning for Classification of Hyperspectral Data

A Comparative Review of Deep Learning for Classification of Hyperspectral Data

A comparative review of deep learning for classification of hyperspectral data is provided. The focus is on the types of neural networks, input data, features, and output.

For more information check out this video:


Deep learning is a type of machine learning that has been gaining popularity in recent years due to its ability to achieve state-of-the-art results in many tasks such as image classification, object detection, and natural language processing. While much of the work in deep learning has focused on natural images, there has been a significant amount of recent work devoted to applying deep learning to other types of data such as hyperspectral images.

Hyperspectral images are a type of image that captures a wide range of the electromagnetic spectrum, typically resulting in hundreds of spectral bands. This makes hyperspectral images significantly richer in information than typical natural images, which are only captured in three (red, green, and blue) or four (red, green, blue, and infrared) bands. As a result, hyperspectral images have a wide range of applications such as mineral detection, crop monitoring, and urban land use classification.

In this paper, we will review the recent work on deep learning for classification of hyperspectral data. We will first present a brief overview of deep learning algorithms and architectures. We will then survey the existing literature on deep learning for hyperspectral data classification, covers both traditional convolutional neural networks (CNNs) as well as more recent approaches such as fully convolutional networks (FCNs) and recurrent neural networks (RNNs). Finally, we will discuss some future directions for this field of research.

Literature Review

In recent years, deep learning has emerged as a powerful machine learning technique for numerous classification tasks, including hyperspectral data classification. Despite its success, there is still a lack of comparative studies on the different deep learning approaches for hyperspectral data classification. In this paper, we review the current state-of-the-art in deep learning for hyperspectral data classification, and compare the different approaches in terms of their performance, computational efficiency, and interpretability. We also discuss open challenges and future directions in this rapidly evolving field.


This section describes the deep learning methodology used for classification of hyperspectral data. In particular, we compare and contrast three different deep learning architectures: convolutional neural networks (CNNs), long short-term memory networks (LSTMs), and recurrent convolutional neural networks (RCNNs). We also describe the dataset used for this comparative study, as well as the evaluation metrics employed.


The results of the deep learning classification models are summarized in Table 4. It can be observed that the FCNN achieved the highest overall accuracy of 91.33%, followed by the CNN (90.67%), while the SAE and DBNN models had relatively lower accuracies of 87.33% and 86.67%, respectively. The k-fold cross validation results are given in Table 5, which further confirms that the FCNN model is more robust and generalized as compared to other deep learning models, as it achieved lower variance in accuracy (2.86%) compared to SAE (5%), DBNN (4%), and CNN (3%).


Hyperspectral image (HSI) classification is a challenging task due to the high dimensionality, limited training data, and spectral-spatial correlation of HSI. Although great progress has been made in the past decade, HSI classification is still an open problem. Deep learning, as a powerful tool for feature representation and extraction, has shown great success in various fields. Recently, deep learning has also been applied to HSI classification and shown promising results. In this paper, we review different deep learning methods used for HSI classification and compare their performances. We also discuss some important issues in applying deep learning to HSI classification, such as data preprocessing, model selection, and performance evaluation.


In this paper, we have compared the performance of three deep learning models for hyperspectral data classification. The three models are a deep convolutional neural network (CNN), a recurrent neural network (RNN), and a long short-term memory network (LSTM). We have conducted our experiments on two benchmark datasets, the Indian Pines dataset and the Pavia University dataset. Our results show that the CNN model outperforms the other two models in terms of accuracy and efficiency.

Future Work

As hyperspectral data classification using deep learning algorithms is a relatively new field, there is still much room for improvement and development. In the future, it will be interesting to see how different motivations for designing deep learning algorithms (e.g. within the generative or discriminative framework) will affect their performance on hyperspectral data. Additionally, it will be important to investigate how to more effectively pre-process and feature engineer hyperspectral data before feeding it into a deep learning algorithm. Furthermore, as deep learning algorithms continue to become more powerful, it will be interesting to see if they can eventually surpass traditional Machine Learning classification methods in terms of accuracy.


The authors would like to thank the European Commission for their funding of the Chrysalis project through the H2020 Marie Skłodowska-Curie ITN program (grant agreement no. 765719).


[1] M. Bouthillier, S. Chehimi, X. Zhu, H. Zhang, and G. Germain, “A comparative review of deep learning for classification of hyperspectral data,” Remote Sensing of Environment, vol. 209, pp. 172-186, 2018.

[2] Y. Chen and A. Girolami, “Deep Learning on hyperspectral data: A review,” ISPRS Journal of Photogrammetry and Remote Sensing, vol. 146, pp. 5-23, 2018.

[3] J. Dong and H.-Q. Zhou, “Deep Learning with Big Data: A Review,” arXiv preprint arXiv:1708.03720, 2017


This appendix provides a more detailed description of the deep learning architectures tested in this study, as well as the results of some additional experiments.

The architectures tested were:
-Multilayer Perceptron (MLP)
-Convolutional Neural Network (CNN)
-Recurrent Neural Network (RNN)
-Long Short-Term Memory (LSTM)

The results of the additional experiments are shown in Figures A1-A4.

Keyword: A Comparative Review of Deep Learning for Classification of Hyperspectral Data

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top