AIRBORNE PARTICLE CLASSIFICATION WITH A COMBINATION OF CHEMICAL-COMPOSITION AND SHAPE INDEX UTILIZING AN ADAPTIVE RESONANCE ARTIFICIAL NEURAL-NETWORK

Airborne particle classification that leads to particle source identif ication is important to both the improvement of the environment and th e protection of public health. In this study, individual airborne part icles were analyzed using a computer-controlled scanning electron micr oscope (CCSEM). It was found that a more accurate particle classificat ion can be obtained when it is based on both the chemical compositions and a shape index of the individual particles compared to one that is based only on the chemical compositions. This study also demonstrated that a newly developed adaptive resonance artificial neural network s ystem (ART2A) has a high potential value in particle classification. T he ART2A system can identify new cluster(s) for the unknown particles and dynamically update the particle class library. Thus, it provides a way to both identify and further investigate new sources for the airb orne particles.