5 May 2010 A framework for evaluating mixture analysis algorithms
Author Affiliations +
In recent years, several sensing devices capable of identifying unknown chemical and biological substances have been commercialized. The success of these devices in analyzing real world samples is dependent on the ability of the on-board identification algorithm to de-convolve spectra of substances that are mixtures. To develop effective de-convolution algorithms, it is critical to characterize the relationship between the spectral features of a substance and its probability of detection within a mixture, as these features may be similar to or overlap with other substances in the mixture and in the library. While it has been recognized that these aspects pose challenges to mixture analysis, a systematic effort to quantify spectral characteristics and their impact, is generally lacking. In this paper, we propose metrics that can be used to quantify these spectral features. Some of these metrics, such as a modification of variance inflation factor, are derived from classical statistical measures used in regression diagnostics. We demonstrate that these metrics can be correlated to the accuracy of the substance's identification in a mixture. We also develop a framework for characterizing mixture analysis algorithms, using these metrics. Experimental results are then provided to show the application of this framework to the evaluation of various algorithms, including one that has been developed for a commercial device. The illustration is based on synthetic mixtures that are created from pure component Raman spectra measured on a portable device.
© (2010) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Sridhar Dasaratha, Sridhar Dasaratha, T. S. Vignesh, T. S. Vignesh, Sarat Shanmukh, Sarat Shanmukh, Malathi Yarra, Malathi Yarra, Edita Botonjic-Sehic, Edita Botonjic-Sehic, James Grassi, James Grassi, Hacene Boudries, Hacene Boudries, Ivan Freeman, Ivan Freeman, Young K. Lee, Young K. Lee, Scott Sutherland, Scott Sutherland, } "A framework for evaluating mixture analysis algorithms", Proc. SPIE 7665, Chemical, Biological, Radiological, Nuclear, and Explosives (CBRNE) Sensing XI, 76650F (5 May 2010); doi: 10.1117/12.849804; https://doi.org/10.1117/12.849804

Back to Top