Improved real-time bio-aerosol classification using Artificial Neural Networks
Abstract. Air contamination has had stronger and stronger impact on everyday life of humans. An increasing number of people are aware of the health problems that may result from inhaling air containing dust, bacteria, pollens or fungi. Society is awaiting anxiously for a system that could inform them in real-time about a real danger that is suspended in the air. The devices, currently available on the market, are able to detect some particles in the air, but cannot classify them by the health threats. Fortunately, a new type of technology is emerging as a really promising solution. Laser based bio-detectors are opening a new era in aerosol research. They are capable of characterizing a great number of individual particles in seconds by analyzing optical scattering and fluorescence characteristics. In this study we demonstrate application of Artificial Neural Network (ANN) to real-time analysis of single particle fluorescence fingerprints. We gathered a total of 114 779 spectra of 48 aerosols. We discuss an entirely new approach to data analysis using decision tree comprising 22 independent neural networks. Applying confusion matrices and ROC analysis the best sets of ANN’s for each group of similar aerosols has been determined. As a result we achieved very high performance of aerosol classification in real-time. We found that for some substances that have characteristic spectra almost each particle can be properly classified. The aerosols with similar spectral characteristics can be classified as a specific cloud with high probability.