基于多波段成像和多任务深度学习的船舶尾气遥测技术研究

船舶尾气光学遥感探测是实现船舶有效监管的重要手段之一，但是目前对污染气体的探测精度和探测能力有限。主要问题在于尾气光学遥感图像存在来自于其他气体和颗粒物的消光干扰，难以建立图像光学特征和污染气体含量的定量关系。我们前期研究发现利用卷积神经网络提取尾气紫外图像特征，通过深度学习消除干扰因素，能预测船用油硫含量。为此，本项目拟结合紫外和热红外两种遥感探测手段，利用卷积神经网络+多任务深度学习，开展多种污染气体和船用油硫含量一体化探测技术研究：1）探寻最优污染气体共有光学特征波段，设计多通道探测方案；2）利用深度神经网络提取并融合多波段尾气图像特征，建立污染气体浓度估测模型；3）通过采集台架实验和实船航行数据验证并优化模型。研究成果为船舶尾气光学遥感探测提供结合人工智能手段的新思路，将船舶尾气光学遥感监测研究再向前推进一步，对于高效实时监管港口船舶，维护国家利益和保护海洋环境具有重要意义。

Optical remote sensing detection is one of the important methods to achieve the effective regulation of ship emissions. However, it has limited accuracy and ability for pollution gases. The main problem lies in the extinction interference caused by other gases and particles in exhaust gases, which makes it difficult to establish the quantitative relationship between the optical features in the image and the concentrations of pollution gases. Our previous research found that the fuel sulfur content can be predicted by using convolution neural network to extract features in ultraviolet image of exhaust gases and eliminating interference factors through deep learning. Therefore, this project plans to carry out the research on integrated detection technology for various pollution gases in ship exhaust and fuel sulfur content by combining ultraviolet and thermal infrared detection methods and using convolution neural network + multi-task deep learning: 1) find the optimal common bands with optical features for pollutant gases; 2) establish the estimation model of pollution gases concentrations by extracting and fusing using multi-band exhaust image features using the deep neural network; 3) verify and optimize the model by collecting data from bench test and real ship navigation. The results will provide a new idea-combining artificial intelligence methods-for optical remote sensing detection of ship exhaust and further advance the research on optical remote sensing regulations of ship exhaust, which is of great significance for efficient real-time supervision of port ships, maintenance of national interests and protection of marine environment.