PAPER Particle Swarms for Feature Extraction of Hyperspectral Data.pdf

1038 IEICE TRANS. INF. SYST., VOL.E90–D, NO.7 JULY 2007 PAPER Particle Swarms for Feature Extraction of Hyperspectral Data Sildomar Takahashi MONTEIRO?a), Student Member and Yukio KOSUGI?, Member SUMMARY This paper presents a novel feature extraction algorithm based on particle swarms for processing hyperspectral imagery data. Parti- cle swarm optimization, originally developed for global optimization over continuous spaces, is extended to deal with the problem of feature extrac- tion. A formulation utilizing two swarms of particles was developed to opti- mize simultaneously a desired performance criterion and the number of se- lected features. Candidate feature sets were evaluated on a regression prob- lem. Artificial neural networks were trained to construct linear and nonlin- ear models of chemical concentration of glucose in soybean crops. Ex- perimental results utilizing real-world hyperspectral datasets demonstrate the viability of the method. The particle swarms-based approach presented superior performance in comparison with conventional feature extraction methods, on both linear and nonlinear models. key words: feature extraction, particle swarm optimization, hyperspectral data, neural networks, principal components analysis 1. Introduction Advances in optical and computational technology have al- lowed the acquisition of an ever-increasing amount of infor- mation from a scene. However, those huge amounts of data represent a challenge for the algorithms to process and ex- tract the relevant information for the desired applications. Despite the wealth of information, the datasets are com- monly plagued by redundant or irrelevant features. In real- world applications, the typical scenario of few data samples in a high-dimensional feature space causes what was termed by Bellman [1] as the curse of dimensionality, referring to the exponential increase in complexity of high-dimensional spaces with the increase in the number of measurements [2]. Hyperspectral im