Surveillance of ship emissions and fuel sulfur content based on imaging detection and multi-task deep learning

Abstract

Shipping makes up the major proportion of global transportation and results in an increasing emission of air pollutants. It accounts for 3.1%, 13%, and 15% of the annual global emissions of CO₂, SO_x, and NO_x, respectively. Hence, effective regulatory measures in line with the International Maritime Organization requirements regarding the fuel sulfur content (FSC) used in emission control areas are essential. An imaging detection approach is proposed to estimate SO₂, CO₂, and NO concentrations of exhaust gas and then calculate FSC based on the estimated gas concentrations. A multi-task deep neural network was used to extract the features from the ultraviolet and thermal infrared images of the exhaust plume. The network was trained to predict various gas concentrations. The results show high prediction accuracy for the remote monitoring of ship emissions.

Keywords: Air pollutants; Convolutional neural network; Fuel sulfur content; Imaging detection; Multi-task learning; Ship exhaust.