<title>Band selection from a hyperspectral data-cube for a real-time multispectral 3CCD camera</title>

Abstract

Given data gr spee( choice This can be done using a real-time muttispectral 3CCD car¡treta, which records a scene with tluee detectors, each accuratcly set to a wavelength by selected opticat filt€rs. This leads to the subject of this papen how ûo rlect tåree optim¡t bards ¡p' hpcrspectrat data to perform a ccrtain task The choice of thesc bands inch¡der lwo aspects, the center wavetengtb, and thc spectral width. A band-selection and band-broadening proccdure has been developed, based on søtistical pattcrn-dopition techniques.We will demonsEate orn proposed band selection algoritln, and present its classification results compared to red-grcer, blue and red-green-near-infrared data for a miliury vehicle in a natur¿l backgrormd a¡d fo¡ surface laid landmines in vegetation.

Multidimensional local spatial autocorrelation measure for integrating spatial and spectral information in hyperspectral image band selection

“…Using the eigenvalues and eigenvectors generated by the CA, a loading factor matrix can be calculated for each spectral band and be used as the classification capability of that band. The Bhattacharyya distance [7] and the Mahalanobis distance [23] have also been widely used for band selection in multispectral image analysis. These measures rank the bands to separate discriminative bands from irrelevant and redundant bands.…”

Sensor band selection for multispectral imaging via average normalized information

“…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.…”

Hyperspectral Data Analysis and Visualization

“…Our research therefore not only looks at the location of the bands but also at the width of the bands. In our previous research (Withagen et al, 2001) a first attempt was made by developing an algorithm that first determines the best locations, in terms of maximizing information content in a limited set of wavelengths, for the bands and than the best width of the bands. This however does not allow for a comparison between broad and narrow bands.…”