@unpublished{,
    author = {Wang, Guanzhong; Ruser, Heinrich; Schade, Julian; Passig, Johannes; Zimmermann, Ralf; Dollinger, Günther; Adam, Thomas},
    title = {Automatic Classification of Ship Emissions using Single-Particle Mass Spectrometry and Deep Learning},
    editor = {},
    booktitle = {},
    series = {},
    journal = {},
    address = {},
    publisher = {},
    edition = {},
    year = {2024},
    isbn = {},
    volume = {},
    number =  {},
    pages = {},
    url = {},
    doi = {},
    keywords = {Aerosol ; ship emission ; air quality},
    abstract = {Air quality control is important for assessing the impact on human health, environment and climate. Single-particle mass spectrometry (SPMS) is a powerful measurement tool to analyse air-transported particle matter (PM) in real-time (Passig 2022). Spectral data of aerosol particles generated by SPMS carry rich information about the chemical composition associated with the sources of the particles, e.g. traffic and ship emissions, biomass burning, etc. (Anders 2023). Accurate classification of aerosol particles is essential to characterize and possibly track these sources.
Common methods to classify PM according to characteristic ion patterns in their mass spectra are based on clustering methods which generally require manual postprocessing and are not suitable for real-time applications. A number of models for automated classification trained on labeled data of ship emissions were proposed recently (Wang 2023, Wang 2024).
In this work, we apply a deep-learning fuzzy convolutional neural network (FCNN) to accurately classify complex mass spectral patterns of single aerosol particles. The hybrid structure of FCNN allows classes of similar but distinctive mass spectra to be distinguished.
The exemplary data used in this study were obtained from a measurement campaign carried out in summer 2023 near Darßer Ort, Germany, a remote location on the coast line of the Baltic Sea, down-wind in approx. 20 km distance from the Kadet channel, the main shipping lane in the Baltic Sea. Considering the relevance of iron-containing particles to ship emissions, we particularly focused on the degree of aging of these particles, as an indication of the distance to the emission source. Therefore, we classified iron-containing particles in more refined classes. Processing > 10,000 mass spectra labeled in 13 classes of maritime origin, classification with high accuracy rates of > 90 % was achieved in less than a second.
To summarize, exploring Deep Learning (DL) approaches like FCNN helps profiling complex MS patterns of aerosol particles and opens up ways to foster rapid air quality assessment and pollution source identification. Ongoing research will further refine and optimize the network architecture, explore more diverse MS patterns, and extend the applicability of the model to different environmental or security-related contexts. This work was funded by dtec.bw – Digitalization and Technology Research Center of the Bundeswehr (LUKAS project). dtec.bw is funded by the European Union – NextGenerationEU.},
    note = {Vortrag bei European Aerosol Conference 2024},
    institution = {Universität der Bundeswehr München, Fakultät für Maschinenbau, MB 6 - Institut für Chemie und Umwelttechnik, Professur: Adam, Thomas},
    
    
}