Selection of optimum training sets for use in pattern recognition analysis of chemical data

Wavelet analysis used for spectral background removal in the determination of glucose from near-infrared single-beam spectra

“…Sample concentrations were designed by a combination of uniform design[17] and subset selection. [18] In the calibration data set, the correlation coefficients between the concentrations of different components ranged from −0.1 to 0.13, while for the prediction data, correlation coefficients between different components concentrations was in the range from −0.27 to 0.21.…”

Pattern recognition methods for automated detection and quantification

“…The backgrounds used were a subset of 5000 selected from the 14,877 available backgrounds that had corresponding altitudes < 3000 ft. The subset selection procedure developed by Carpenter and Small 78 was used to select the 5000 background spectra. 137 Cs collected from a calibration source is scaled by a randomly selected constant (e.g., 0.47 in the pictured example) and added to a randomly selected background spectrum collected from the air.…”

“…In addition to the test sets described above, a database of >500,000 airborne background interferograms collected with the same instrumentation and aircraft described above were available for use in the synthesis of ammonia-active interferograms and as the inactive data class in training the ammonia classifiers. The subset selection algorithm developed by Carpenter and Small 78 was used to obtain representative samplings of this pool for use in the data analysis steps described below.…”

Signal processing and pattern recognition methods for the remote, airborne detection of radioisotopes by gamma-ray spectroscopy

“…After being generated separately the seven groups of 5,000 synthetic actives were combined for form a total set of 35,000 active patterns. The subset selection algorithm developed by Carpenter and Small 32,77,97 was then used to obtain a representative set of 16,000 active patterns that was employed in training the classifiers. The training set inactive class was selected from the database of in-flight background spectra summarized in Chapter 4.…”

“…To incorporate these sources of variation, the pure-component spectrum was randomly scaled to account for variation in signal intensities in the field and then added to a randomly selected background spectrum to generate a synthetic active spectrum. For each synthesized spectrum, the corresponding background spectrum was taken from a group of 5,000 field backgrounds.These spectra were selected as representative from the background pool ofTable 4-1 by use of the algorithm developed by Carpenter and Small 77,97. The pool of 5,000 backgrounds did not overlap with those in the inactive class of the training set.…”