NCJ Number
249454
Journal
Talanta Volume: 132 Dated: January 2015 Pages: 182-190
Date Published
January 2015
Length
9 pages
Annotation
This study used a two-step procedure to develop search prefilters (i.e., principal component models) that differentiate between similar but not identical IR spectra of automotive clear coats.
Abstract
Clear coat searches of the infrared (IR) spectral library of the paint data query (PDQ) forensic database often generate an unusable number of hits that span multiple manufacturers, assembly plants, and years. To improve the accuracy of the hit list, pattern recognition methods have been used to develop search prefilters that differentiate between similar but non-identical IR spectra of clear coats on the basis of manufacturer (e.g., General Motors, Ford, Chrysler) or assembly plant. The current study first used the discrete wavelet transform to decompose each IR spectrum into wavelet coefficients to enhance subtle but significant features in the spectral data. Second, a genetic algorithm for IR spectral pattern recognition was used to identify wavelet coefficients characteristic of the manufacturer or assembly plant of the vehicle. Even in challenging trials where the paint samples evaluated were all from the same manufacturer (General Motors) within a limited production year range (2000-2006), the respective assembly plant of the vehicle was correctly identified. Search prefilters for identifying assembly plants were successfully validated using 10 blind samples provided by the Royal Canadian Mounted Police (RCMP) as part of a study to populate PDQ to current production years; and the search prefilter to discriminate among automobile manufacturers was successfully validated using IR spectra obtained directly from the PDQ database. (Publisher abstract modified)