Feature Selection for Anomaly Detection in Hyperspectral Data

Over the past decade, use of hyperspectral imagery has been intensively investigated for agricultural product inspection, since it introduces a new noninvasive machine-vision method that gives a very accurate inspection rate. The spectral information in hyperspectral data uniquely characterizes and identifies the chemical and/or physical properties of the constituent parts of an agricultural product that are useful for product inspection. One of the main problems in using these high-dimensional data is that there are often not enough training samples. This book, therefore, provides novel feature selection algorithms to effectively reduce the dimensionality of hyperspectral data. Experimental results comparing the proposed algorithms to other well-known feature selection algorithms are presented for two case studies in chicken carcass inspection. This book provides insightful discussions on feature selection for hyperspectral data for specific food safety applications and should be especially useful to engineers and scientists who are interested in pattern recognition, hyperspectral data processing, food safety research, and data mining.