Measurement of two-dimensional angle-resolved optical scattering (TAOS) patterns is an attractive technique for detecting and characterizing micron-sized airborne particles. In general, the interpretation of these patterns and the retrieval of the particle refractive index, shape or size alone, are difficult problems. By reformulating the problem in statistical learning terms, a solution is proposed herewith: rather than identifying airborne particles from their scattering patterns, TAOS patterns themselves are classified through a learning machine, where feature extraction interacts with multivariate statistical analysis. Feature extraction relies on spectrum enhancement, which includes the discrete cosine FOURIER transform and non-linear operations. Multivariate statistical analysis includes compu- tation of the principal components and supervised training, based on the maximization of a suitable figure of merit. All algorithms have been combined together to analyze TAOS patterns, organize feature vectors, design classification experiments, carry out supervised training, assign unknown patterns to classes, and fuse information from different training and recognition experiments. The algorithms have been tested on a data set with more than 3000 TAOS patterns. The parameters that control the algorithms at different stages have been allowed to vary within suitable bounds and are optimized to some extent. Classification has been targeted at discriminating aerosolized Bacillus subtilis particles, a simulant of anthrax, from atmospheric aerosol particles and interfering particles, like diesel soot. By assuming that all training and recognition patterns come from the respective reference materials only, the most satisfactory classification result corresponds to 20% false negatives from B. subtilis particles and <11% false positives from all other aerosol particles. The most effective operations have consisted of thresholding TAOS patterns in order to reject defective ones, and forming training sets from three or four pattern classes. The presented automated classification method may be adapted into a real-time operation technique, capable of detecting and characterizing micron-sized airborne particles.

Crosta, G., Pan, Y., Aptowicz, K., Pinnick, R., Chang, R., Videen, G. (2013). Automated classification of single airborne particles from two-dimensional angle-resolved optical scattering (TAOS) patterns by non-linear filtering. JOURNAL OF QUANTITATIVE SPECTROSCOPY & RADIATIVE TRANSFER, 131, 215-233 [10.1016/j.jqsrt.2013.06.009].

Automated classification of single airborne particles from two-dimensional angle-resolved optical scattering (TAOS) patterns by non-linear filtering

CROSTA, GIOVANNI FRANCO FILIPPO;
2013

Abstract

Measurement of two-dimensional angle-resolved optical scattering (TAOS) patterns is an attractive technique for detecting and characterizing micron-sized airborne particles. In general, the interpretation of these patterns and the retrieval of the particle refractive index, shape or size alone, are difficult problems. By reformulating the problem in statistical learning terms, a solution is proposed herewith: rather than identifying airborne particles from their scattering patterns, TAOS patterns themselves are classified through a learning machine, where feature extraction interacts with multivariate statistical analysis. Feature extraction relies on spectrum enhancement, which includes the discrete cosine FOURIER transform and non-linear operations. Multivariate statistical analysis includes compu- tation of the principal components and supervised training, based on the maximization of a suitable figure of merit. All algorithms have been combined together to analyze TAOS patterns, organize feature vectors, design classification experiments, carry out supervised training, assign unknown patterns to classes, and fuse information from different training and recognition experiments. The algorithms have been tested on a data set with more than 3000 TAOS patterns. The parameters that control the algorithms at different stages have been allowed to vary within suitable bounds and are optimized to some extent. Classification has been targeted at discriminating aerosolized Bacillus subtilis particles, a simulant of anthrax, from atmospheric aerosol particles and interfering particles, like diesel soot. By assuming that all training and recognition patterns come from the respective reference materials only, the most satisfactory classification result corresponds to 20% false negatives from B. subtilis particles and <11% false positives from all other aerosol particles. The most effective operations have consisted of thresholding TAOS patterns in order to reject defective ones, and forming training sets from three or four pattern classes. The presented automated classification method may be adapted into a real-time operation technique, capable of detecting and characterizing micron-sized airborne particles.
Articolo in rivista - Articolo scientifico
Angle-resolved optical scattering; Machine learning; Airborne particle; Scattering pattern; Feature extraction; Linear classifier; Supervised training; Bacillus subtilis; Outdoor sampling
English
2013
131
215
233
reserved
Crosta, G., Pan, Y., Aptowicz, K., Pinnick, R., Chang, R., Videen, G. (2013). Automated classification of single airborne particles from two-dimensional angle-resolved optical scattering (TAOS) patterns by non-linear filtering. JOURNAL OF QUANTITATIVE SPECTROSCOPY & RADIATIVE TRANSFER, 131, 215-233 [10.1016/j.jqsrt.2013.06.009].
File in questo prodotto:
File Dimensione Formato  
Crosta-Pan_2013.pdf

Solo gestori archivio

Tipologia di allegato: Other attachments
Dimensione 3.62 MB
Formato Adobe PDF
3.62 MB Adobe PDF   Visualizza/Apri   Richiedi una copia

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/10281/47995
Citazioni
  • Scopus 10
  • ???jsp.display-item.citation.isi??? 8
Social impact