<title>Information theory-based band selection for multispectral systems</title>

Abstract

This paper describes a methodology we have developed for wavelength band selection. This methodology combines an information theory-based criterion for selection with a genetic algorithm for searching for a nearoptima1 solution. We have applied this methodology to 302 material spectra in the Nonconventional Exploitation Factors (NEF) database to determine the band locations for 7, 15, 30, and 60-band sets that permit the best material separation. These optimal band sets were also evaluated in terms of their utility related to anomaly/target detection using multiband images generated from a Hyperspectral Digital Imagery Collection Experiment (HYDICE) image cube. The optimal band locations and their corresponding entropies are given in this paper. Also presented are the anomaly/target detection results obtained from using these optimal band sets.

Sensor band selection for multispectral imaging via average normalized information

“…High entropy values indicate that each value is equally likely to occur and are thus not indicative of the particular material. Low entropy values on the other hand can be attributed to the existence of a persistent pattern [4,16,18,22]. Hence, the preferred features should have small joint entropy values H jnt (see Sect.…”

“…Low entropy values correspond to low uncertainty and thus bands with small entropy are considered good feature bands. Bassett and Shen [4] use entropy to measure the difference between different classes. To maximize that difference, they select the bands with the biggest total entropy.…”

“…Information theory has also been used in feature bands selection [3,4]. Entropy is interpreted as a measure of stability of each individual wavelength band in [3].…”

“…As multispectral/hyperspectral imaging is becoming more widely used, many researchers and practitioners are focusing on real-time classification in multispectral imagery [7] and automated wavelength band selection [3,4,6,9,19,26,30]. Most of the wavelength selection methods proposed so far usually involve two separate tasks: (a) the task of selecting the wavelength bands which carry significant information, known as feature bands selection; and (b) the task of eliminating those feature bands that contribute redundant information, known as redundancy reduction.…”

Multispectral texture characterization: application to computer aided diagnosis on prostatic tissue images

“…Several approaches have been investigated by looking into how to remove information re-dundancy resulting from highly correlated bands [33,[37][38][39][40]. Most of the methods usually involve two separate tasks: (a) selecting the bands that can indicate the particular material well, feature bands selection; and (b) removing the feature bands contributing redundant information, redundancy reduction [41].…”