Using a Convolutional Neural Network and Mid-Infrared Spectral Images to Predict the Carbon Dioxide Content of Ship Exhaust
Round 1
Reviewer 1 Report
The authors developed a method for predicting the CO2 content of ship exhaust that uses a convolutional neural network and mid-infrared spectral images. .I think this is a very interesting research topic.
I believe that before publishing this article, the following issues should be addressed and corresponding modifications should be madel.
1)Fig 6 Structural diagram of the residual block is a classic structural diagram and should indicate the source of literature.
2)There is a lack of ablation study in the analysis of prediction results using neural networks.
3)In the analysis of experimental results, there is still a lack of general quantitative evaluation indicators, and only three conventional accuracy indicators are provided, as seen in fig.7.
Author Response
Dear Reviewer:
On behalf of my co-authors, we sincerely thank you for giving us the opportunity to revise our manuscript. We appreciate reviewer’s constructive comments and suggestions on our manuscript. We have studied them carefully and have made corrections that we hope will meet the expectations. The main corrections and responses to the reviewer's comments in the paper are attached, please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
This manuscript proposed a novel deep learning method for predicting carbon dioxide content of ship exhaust based on mid-infrared spectral images, where convolutional neural network (CNN) was employed for the task of interest. First, a bench experiment was performed to synchronously obtain mid-wave infrared spectral images of the ship exhaust plume and true values for the CO2 concentration from the online monitoring of eight spectral channels. Then, a deep CNN architecture based on ResNet50 was established for predicting CO2 content. Finallly, the experimental results validate the capability of proposed method, with satisfactory results. Overall, the topic of this research is interesting, and the manuscript was well organised and written. The detailed comments are summarised as follows.
1. The main innovation and contribution of this research should be clearly clarified in abstract and introduction.
2. Please broaden and update literature review on CNN or deep learning to demonstrate its excellent capacity to resolve practical prediction problems. E.g. Torsional capacity evaluation of RC beams using an improved bird swarm algorithm optimised 2D convolutional neural network; Automated damage diagnosis of concrete jack arch beam using optimized deep stacked autoencoders and multi-sensor fusion.
3. ResNet50 is a pretrained CNN, which is used to classify 1000 categories. Accordingly, transfer learning should be employed to convert ResNet to the model for this application. Please add more information on how to transfer the model.
4. The performance of proposed network is heavily dependent on the setting of hyperparameters. How did the authors tune/optimise the network parameters to achieve the best prediction accuracy in this research?
5. A comparison with other commonly used method in the literature is suggested.
6. More future research should be included in conclusion part.
Acceptable
Author Response
Dear Reviewer:
On behalf of my co-authors, we sincerely thank you for giving us the opportunity to revise our manuscript. We appreciate reviewer’s constructive comments and suggestions on our manuscript. We have studied them carefully and have made corrections that we hope will meet the expectations. The main corrections and responses to the reviewer's comments in the paper are attached, please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
The publication provides interesting experiment for the evaluation of exhaust gas CO2 concentration. How ever improvements to the manuscript should be made.
1. It is not clear what authors mean with the statement „maximum exhaust gas treatment capacity of 30,000 kg/h“ If this means that the maximum flow of flue gas is 30 000 kg/h than it should be clearly stated.
2. It should be explained if the measurements were done in the engine laboratory conditions using marine engine or on a ship and this should be reflected in the conclusions.
3. It is also not clear what is meant by “The different stable operating conditions were then used to eliminate the effect of the engine power on the inversion results.“. Engine load directly influences the flow of exhaust gas, as well as the gas temperature and even concentrations of pollutants in exhaust gas. Additional explanation needed.
4. In the description of measurement, it is stated that temperature, pressure and even flow and speed of exhaust gas were measured. But it is not clear if these parameters were used in the training of the network. In the training results only concentration of CO2 was provided.
5. Figure 4 is of poor quality and provide limited information about the experiment to the reader.
6. In description of experiment it should be noted what fuel was used in the engine, as marine fuel composition (contrary to automotive diesel fuel) can differ among different bunkers, including the differences in carbon content and generation of other pollutants.
7. In figure 7 it is not clear what abscissa axis present.
8. The goal of the study should be more clearly defined. The need co control CO2 is controlled and evaluated trough the evaluation of used fuel. It can also be evaluated from measurement of emissions, but for concentration of CO2 and flow of exhaust gas is needed to obtain emissions in grams. in addition, engine load should be simultaneously evaluated if gCO2/kWh is needed. How ever applying this method only provides concentration of CO2 which on its own does not provide data for any CO2 benchmark of a ship. CO2 measurement can, if coupled with measurements of SO2, provide data ship fuel sulphur content which is strictly regulated. But if only CO2 concentration is measured it should be defined what is the goal of such monitoring.
Author Response
Dear Reviewer:
On behalf of my co-authors, we sincerely thank you for giving us the opportunity to revise our manuscript. We appreciate reviewer’s constructive comments and suggestions on our manuscript. We have studied them carefully and have made corrections that we hope will meet the expectations. The main corrections and responses to the reviewer's comments in the paper are attached, please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
All the technical issues have been well addressed by the authors. I suggest this revised version can be accepted for publication.
Reviewer 3 Report
The topic is interesting and well explained. Looking forward to see more of your work in this dirrection.
