A memetic algorithm for generating spectral indices for remotely sensed imagery

Issue Date

10-2019

Abstract

Spectral indices are formulas that integrate different wavelengths, or bands, of the electromagnetic spectrum to accentuate the abundance of various features of interest such as vegetation, burned areas, man-made, water, and geological features. Given its numerous applications, research on the development of spectral indices is a worthwhile undertaking. New spectral indices could be created as improvements over existing ones in aspects such as separability and sensitivity. New spectral indices can be developed for other features not yet covered by any spectral index. The problem of developing spectral indices was posed as a search problem. The search space consists of all possible spectral indices. A spectral index was viewed as a combination of various spectral bands, mathematical operators, and numerical coefficients. Therefore, search algorithms could be used to generate spectral indices. For this study, the memetic algorithm was utilized. Specifically, a memetic variant of genetic programming was developed. This was achieved by augmenting the genetic programming algorithm with the simulated annealing algorithm. The developed algorithm was used to generate spectral indices for vegetation and built-up areas. Training points derived from Landsat 8 imagery was used as the input. The quality of the generated spectral indices was measured using two metrics: Silhouette Score and the Jeffries-Matusita distance. Indices for vegetation and built-up areas were developed: namely Memetic Genetic Programming Vegetation Index (MGPVI) and Memetic Genetic Programming Built-up Index (MGPBI). MGPVI was compared with other vegetation indices and MGPBI was compared with other built-up indices. Both generated indices outperformed their competing indices in terms of the earlier mentioned metrics.

Source or Periodical Title

40th Asian Conference on Remote Sensing, ACRS 2019: "Progress of Remote Sensing Technology for Smart Future"

Page

1-10

Document Type

Conference Paper

Physical Description

illustrations; diagrams; tables; references

Language

English

Subject

Feature Extraction, Memetic Computing, Search Algorithms, Spectral Index

Digital Copy

yes

Link to Full Text

Share

COinS