Automated classification of single airborne particles from two-dimension, angle-resolved optical scattering (TAOS) patterns

Abstract

Two-dimension, angle-resolved optical scattering (TAOS) is an experimental technique by which patterns of LASER light intensity scattered by single (micrometer or sub-micrometer sized) airborne particles are collected. In the past 10 years TAOS instrumentation has evolved from laboratory prototypes to field-deployable equipment; patterns are collected by the thousands during indoor or outdoor sampling in short times. Although comparison between experimental and computed scattering patterns has been carried out extensively, there is no satisfactory way to relate a given pattern to the particle it comes from. This paper reports about the ongoing development and implementation of a method which is aimed at classifying patterns, rather than identifying original particles. A machine learning algorithm includes the extraction of morphological features and their multivariate statistical analysis. A classifier is trained and validated in a supervised mode, by relying on patterns from known materials. Then the tuned classifier is applied to the recognition of patterns of unknown origin.