1. Discrimination of airborne material particles from light scattering (TAOS) patterns
- Author
-
Kevin B. Aptowicz, Yong-Le Pan, Giovanni F. Crosta, Richard K. Chang, Gorden Videen, Southern, SO, Crosta, G, Pan, Y, Videen, G, Aptowicz, K, and Chang, R
- Subjects
FIS/06 - FISICA PER IL SISTEMA TERRA E PER IL MEZZO CIRCUMTERRESTRE ,Feature extraction ,Linear classifier ,MED/46 - SCIENZE TECNICHE DI MEDICINA DI LABORATORIO ,ING-INF/05 - SISTEMI DI ELABORAZIONE DELLE INFORMAZIONI ,Light scattering ,symbols.namesake ,Optics ,Machine learning ,False positive paradox ,Bacillus subtili ,Physics ,Scattering ,business.industry ,Scattering pattern ,Supervised training ,Pattern recognition ,Angle-resolved optical scattering ,MAT/06 - PROBABILITA E STATISTICA MATEMATICA ,FIS/01 - FISICA SPERIMENTALE ,Fourier transform ,Outdoor sampling ,symbols ,ING-INF/04 - AUTOMATICA ,Monochromatic color ,Artificial intelligence ,business ,Classifier (UML) ,Airborne particle - Abstract
Two-dimensional angle-resolved optical scattering (TAOS) is an experimental method which collects the intensity pattern of monochromatic light scattered by a single, micron-sized airborne particle. In general, the interpretation of these patterns and the retrieval of the particle refractive index, shape or size alone, are difficult problems. The solution proposed herewith relies on a learning machine ( LM ): rather than identifying airborne particles from their scattering patterns, TAOS patterns themselves are classified. The LM consists of two interacting modules: a feature extraction module and a linear classifier. Feature extraction relies on spectrum enhancement , which includes the discrete cosine Fourier transform and non-linear operations. Linear classification relies on multivariate statistical analysis. Interaction enables supervised training of the LM . The application described in this article aims at discriminating the TAOS patterns of single bacterial spores ( Bacillus subtilis ) from patterns of atmospheric aerosol and diesel soot particles. The latter are known to interfere with the detection of bacterial spores. Classification has been applied to a data set with more than 3000 TAOS patterns from various materials. Some classification experiments are described, where the size of training sets has been varied as well as many other parameters which control the classifier. By assuming all training and recognition patterns to come from the respective reference materials only, the most satisfactory classification result corresponds to ≈ 20% false negatives from Bacillus subtilis particles and ≤ 11% false positives from environmental and diesel particles.
- Published
- 2013
- Full Text
- View/download PDF