Journal of Spectral Imaging,   Volume 11   Article ID a2   (2022)

Peer reviewed Paper

A semi-supervised cycle-GAN neural network for hyperspectral image classification with minimum noise fraction

  • Tatireddy Subba Reddy
  • Jonnadula Harikiran  
Research Scholar, School of Computer Science and Engineering, VIT-AP University, Amaravathi, Andhra Pradesh-522237, India and Assistant Professor, Department of Computer Science and Engineering, B V Raju Institute of Technology, Narsapur, Medak, Telangana, India
[email protected]
 Search for papers by this author
 Corresponding Author
Associate Professor, School of Computer Science and Engineering, VIT-AP University, Amaravthi, Andhra Pradesh-522237, India
[email protected]
 Search for papers by this author

Hyperspectral imaging (HSI) is a popular mode of remote sensing imaging that collects data beyond the visible spectrum. Many classification techniques have been developed in recent years, since classification is the most crucial task in hyperspectral image processing. Furthermore, extracting features from hyperspectral images is challenging in many scenarios. The semi-supervised classification of HSI is motivated by the Cycle-GAN method that has been proposed in this research paper. Since the proposed HSI classification method is semi-supervised, it makes extensive use of the labelled samples, which are short and have numerous unlabelled images. The research is carried out in two phases. First, to extract the spectral–spatial features, the minimum noise fraction is adopted. And, second, the classification of the semi-supervised method is done by the cycle-GANs. Subsequently, the proposed architecture is implemented on three standard hyperspectral dataset methods. As a result, the performance comparison is carried out in the same field as state-of-the-art approaches. The obtained results successfully demonstrate the supremacy of the proposed technique in the classification of HSI.

Keywords: feature extraction, hyperspectral image, cycle-GANs, semi-supervised classification, minimum noise fraction (MNF), AVIRIS sensors, rolling guidance filter (RGF), Indian Pines, transductive support vector machine (TSVM), convolutional neural networks (CNN)




Abstract Views: