Journal of Spectral Imaging,   Volume 9   Article ID a14   (2020)

Peer reviewed Paper

Spectral similarity algorithm-based image classification for oil spill mapping of hyperspectral datasets

  • Deepthi
  • Tessamma Thomas
Department of Electronics, Cochin University of Science and Technology, Kochi, Kerala, India 682022

 Search for papers by this author
 Corresponding Author
Department of Electronics, Cochin University of Science and Technology, Kochi, Kerala, India 682022
[email protected]
 Search for papers by this author

In remote sensing, the compositional information of part of the earth’s surface is statistically evaluated by comparing known field or library spectra with the unknown image spectra, known as spectral matching or spectral similarity analysis. In this research, hybrid spectral similarity algorithms developed based on chi-square distance (CHI or χ2) are used to retrieve useful information from the Hyperion hyperspectral oil spill image covering the area near Liaodong Bay of the Bohai Sea, China. In order to evaluate the discriminability of spectral similarity algorithms, a pixel-level matching is carried out between the reference vectors, viz. Oil Slick (O), Sheen (H), Sea Water (S) and Ship Track (T), collected visually from known areas in the image. The hybrid spectral similarity algorithms are statistically assessed for their performance using the spectral discriminatory measures (i) relative spectral discriminatory power (RSDPW), (ii) relative spectral discriminatory probability (RSDPB) and (iii) relative spectral discriminatory entropy (RSDE). Additionally, the selected hybrid algorithms are used on the Hyperion image subset to perform a pixel-based classification. Classification results revealed that the CHI-based hybrid algorithms performed better than all other hybrid spectral similarity methods. Therefore, the CHI-based hybrid algorithms demonstrated their superior spectral discrimination capacity to classify marine spectral classes for oil spill mapping from the hyperspectral dataset.

Keywords: chi-square distance, hybrid spectral similarity measures, hyperspectral image, oil spill, overall accuracy, spectral discriminatory ­measures

Metrics

Downloads:

114

Full-Text HTML Views:

1

Abstract Views:

397