Home Articles Innovative genetic algorithm for hyperspectral image classification

Innovative genetic algorithm for hyperspectral image classification

Abstract

Mr. Chaichoke Vaiphasa
Natural Resources Department,
International Institute for Geo-Information Science and Earth Observation (ITC),
the Netherlands
E-MAIL: [email protected], [email protected]

Abstract
This paper reports the performance of newly developed genetic algorithms for hyperspectral image classification. The proposed algorithms are a combination of genetic search algorithm (GA) and commonly known remote sensing supervised classifiers. The algorithms use the law of natural selection to search for the optimal combination of hyperspectral image layers that gives highest total classification accuracy by which every image band is treated as if it is a single gene of an imaginary living organism. The HYMAP hyperspectral data of salt marsh wetland of Schiermonnikoog Island, the Netherlands is used for testing the performance of this combined classifier. The study area is thoroughly covered by 18 types of salt marsh vegetation species. Even though the spectral signatures of 18 salt marsh species are alike, the proposed classifier gains high classification accuracy.