Deep learning based aerosol particle classification for the detection of ship emissions

Area of Science:

• Environmental Science
• Analytical Chemistry
• Artificial Intelligence

Background:

• International Maritime Organization (IMO) established Sulfur Emission Control Areas (SECA) to mitigate shipping's air pollution impact.
• Ships in SECA must use low-sulfur fuel or exhaust gas cleaning systems (scrubbers).
• Conventional monitoring methods lack real-time data, detection range, and source attribution capabilities.

Purpose of the Study:

• To develop an advanced monitoring system for real-time identification of ship emissions.
• To overcome limitations of traditional methods in detecting and attributing air pollution sources.
• To specifically identify ships combusting heavy fuel oil (HFO) within SECA.

Main Methods:

• Integration of single-particle mass spectrometry (SPMS) with deep learning (convolutional neural network - CNN).
• SPMS for detailed chemical composition analysis of individual aerosol particles.
• CNN for automated classification of aerosol particles (92% accuracy across 13 classes).
• Correlation of CNN predictions with wind data and Automatic Identification System (AIS) ship trajectories.

Main Results:

• The system accurately classifies aerosol particles from multiple sources in real-time.
• Successfully identified unique particles rich in vanadium, nickel, and iron, indicative of HFO combustion.
• Detected 21 ships using HFO within a 1.3 km range during a one-week monitoring period.

Conclusions:

• The proposed SPMS-deep learning system offers a robust solution for real-time ship emission monitoring.
• Enables rapid and reliable detection of heavy fuel oil combustion, crucial for enforcing emission regulations.
• Enhances capabilities for pollution source attribution and air quality management in maritime environments.