Ship Detection in Maritime Scenes under Adverse Weather Conditions

Response to Reviewer 1 Comments

1. Summary

Thank you for your letter and insightful comments on our manuscript entitled “Ship Detection in Maritime Scenes Under Adverse Weather Conditions”. These comments not only greatly assist us in improving our paper, but also have important guiding significance for our subsequent research. We have studied the comments carefully and have made corrections to improve our paper. We appreciate for your warm work earnestly, and hope these corrections will meet with approval.

2. Questions for General Evaluation

Reviewer’s Evaluation

Response and Revisions

Does the introduction provide sufficient background and include all relevant references?

Can be improved

Are all the cited references relevant to the research?

Can be improved

Is the research design appropriate?

Must be improved

Are the methods adequately described?

Must be improved

Are the results clearly presented?

Must be improved

Are the conclusions supported by the results?

Can be improved

3. Point-by-point response to Comments and Suggestions for Authors

Comments 1: ‘To our knowledge, no works focus on utilizing point cloud data for detecting large objects.’ It is not appropriate to say that, because many studies are based on airborne LiDAR and use deep learning methods to study large buildings.

Response 1: Thank you for your reminder. After reviewing the literature, we have revised this statement. For detailed content, refer to the first sentence of the first paragraph in Section 5.4, which is marked in yellow.

Comments 2: In figure 3, compayellowwith figure a, figure b does not seem to have changed except for some additional noise points, while in figure 4, the point clouds on rainy days are significantly sparse. Under adverse weather conditions, especially in thick fog and haze weather, the penetration ability of laser LiDAR decreases significantly. We have used Riegl-VZ1000 three-dimensional laser scanner to test under dense fog conditions, and its penetration power is not even more than 30m. In rainy and snowy days, except for some noise points, the scanning point cloud is almost unchanged. Therefore, it is suggested that the actual measured data can be given, and the static features (such as large buildings, port facilities, or ships docked for a long time, etc.) should be selected and compared with the sunny weather under the conditions of fog, haze, rain, snow and so on. And the comparison with the simulation to verify the reliability of the simulation data. If the difference is large, the simulation model needs to be modified.

Response 2: Thank you for your suggestions. The point clouds were collected using a ship-mounted LiDAR sensor specifically designed for maritime scenes, with a scanning range of one nautical mile. Due to inherent noise in the sensor, the collected data contain noise under both clear and adverse weather conditions. In experiments, point cloud data can be captured in both foggy and rainy conditions. The air during foggy conditions contains impurities, adding noise to the scanned point clouds. The laser beams from the LiDAR can penetrate transparent materials, hence they also penetrate rain; however, the scattering and refraction caused by water droplets create disturbances around the targets and reduce the number of data points collected. Based on your valuable suggestions, we have added 3D visualization and simulated point cloud data validation. Detailed content can be found in Figures 3 and Figures 4, and Sections 3.2.2 and 3.3.2, which are marked in yellow. Thank you once again for your suggestions.

Comments 3: The core of this paper is the ship recognition network model based on deep learning, Dual-branch Sparse 3D Feature Net and Multi-scale 2D CNN Module. The method is described in detail, but the basis of network construction and innovation need to be highlighted.

Response 3: Thank you for your suggestions. Our network extracts point cloud features using 3D sparse convolution, while Second is a representative network that utilizes sparse convolution for detecting small targets. The revised manuscript highlights the foundation of network construction and innovation. Specific details can be found in Section 1, from the fourth to the ninth sentences of the fourth paragraph, which are highlighted with a yellow background. Thank you once again for your suggestions.

4. Response to Comments on the Quality of English Language

Response: Thank you very much for your suggestions. We have meticulously checked the article's grammar multiple times to ensure its correctness and have made several revisions accordingly. We sought feedback from several native speakers to enhance the clarity, fluency, and professionalism of the article.

5.Questions for General Evaluation

Questions 1: Does the introduction provide sufficient background and include all relevant references?

Response 1: The revised manuscript has updated the introduction and added two references to provide a more comprehensive background. Specific details can be found in Section1, from the first to the third sentence of the third paragraph, which are highlighted with a yellow background.:

1. Application Research of Ship Overload Identification Algorithm Based on LiDAR Point Cloud

2. Preliminary Study for Motion Pose of Inshore Ships Based on Point Cloud: Estimation of Ship Berthing Angle.

Questions 2: Are all the cited references relevant to the research?

Response 2: We carefully reviewed each reference, removing those that did not fit the article and adding literature relevant to the field.

Questions 3: Is the research design appropriate?

Response 3: We revised the rainy day point cloud simulation process, adding Gaussian distribution filtering to the point clouds. More detailed information can be found in Section 3.3.1, from the fifth sentence to the end of the last paragraph, highlighted with a yellow background.

Questions 4: Are the methods adequately described?

Response 4: Our manuscript has revised Figure 2 and added Sections 3.2.2 and 3.3.2 to describe the simulation methods.

Questions 5: Are the results clearly presented?

Response 5: Our manuscript has revised Figures 3, Figures 4, and Table 1 to clearly display the simulation and detection results.

Questions 6: Are the conclusions supported by the results?

Questions 6: Our manuscript includes an addition to the third paragraph of Section 5.4 to describe the detection results.

Response to Reviewer 2 Comments

1. Summary

Thank you for your letter and insightful comments on our manuscript entitled “Ship Detection in Maritime Scenes Under Adverse Weather Conditions”. These comments not only greatly assist us in improving our paper, but also have important guiding significance for our subsequent research. We have studied the comments carefully and have made corrections to improve our paper. We appreciate for your warm work earnestly, and hope these corrections will meet with approval.

2. Questions for General Evaluation

Reviewer’s Evaluation

Response and Revisions

Does the introduction provide sufficient background and include all relevant references?

Can be improved

Are all the cited references relevant to the research?

Can be improved

Is the research design appropriate?

Yes

Are the methods adequately described?

Can be improved

Are the results clearly presented?

Yes

Are the conclusions supported by the results?

Yes

3. Point-by-point response to Comments and Suggestions for Authors

Comments 1: Validation Against Actual Adverse Weather Data: Comparing simulated data with actual adverse weather data could validate the simulation's accuracy.

Response 1: Thank you for your helpful suggestions. In the manuscript, Sections 3.2.2 and 3.3.2, marked with a yellow background, compare simulated data, actual adverse weather data, and clear weather data to verify the accuracy of the simulation.

Comments 2: Visual Illustrations: Visual comparisons between simulated and real adverse weather conditions would demonstrate the simulation's effectiveness.

Response 2: Thank you for your suggestions. We have modified Figures 3 and Figures 4 to more intuitively compare the simulated data with actual adverse weather data.

Comments 3: Dataset Diversity: A detailed discussion on the dataset diversity, including ship types and weather conditions, would enhance the assessment of the model's applicability.

Response 3: Thank you for your suggestions. The discussion on the diversity of ship types is marked with a yellow background in the second paragraph of Section 5.4. The discussion on weather diversity is highlighted from the eighth sentence to the end of the third paragraph of Section 5.4 with a yellow background.

Comments 4: Performance Benchmarks: Detailed benchmarks against existing detectors would offer clearer insights into the proposed method's strengths.

Response 4: Thank you for your suggestions. Since our ship detection network utilizes a voxel-based method for point cloud feature extraction, we have selected SECOND, a network designed for small target detection, and adjusted its parameters to serve as benchmarks. Detailed information is located in Section 5.4, specifically in the second to the third sentence of the first paragraph, highlighted with a yellow background.

Comments 5: Model Limitations: Exploring the model's limitations and assumptions could provide insights into potential practical implications.

Response 5: Thanks for your valuable suggestions. The revised script discusses the limitations of our ship detector from the perspective of the range of input data. More specific details are provided in the last paragraph of Section 6, which with yellow background. Thanks for your valuable suggestions.

Comments 6: Technical Summaries: Simplified summaries would make the paper more accessible to a broader audience.

Response 6: Thank you for your suggestions, which have helped improve the quality of the manuscript. We have simplified Section 6 to more directly discuss point cloud simulation and network architecture, enhancing the readability of the paper. More details can be found in the first and second paragraphs of Section 6, marked with a yellow background.

4. Response to Comments on the Quality of English Language

Response: Thank you very much for your suggestions. We have meticulously checked the article's grammar multiple times to ensure its correctness and have made several revisions accordingly. We sought feedback from several native speakers to enhance the clarity, fluency, and professionalism of the article.

5.Questions for General Evaluation

Questions 1: Does the introduction provide sufficient background and include all relevant references?

Response 1: The revised manuscript has updated the introduction and added two references to provide a more comprehensive background. Specific details can be found in Section1, from the first to the third sentence of the third paragraph, which are highlighted with a yellow background.:

1. Application Research of Ship Overload Identification Algorithm Based on LiDAR Point Cloud

2. Preliminary Study for Motion Pose of Inshore Ships Based on Point Cloud: Estimation of Ship Berthing Angle.

Questions 2: Are all the cited references relevant to the research?

Response 2: We carefully reviewed each reference, removing those that did not fit the article and adding literature relevant to the field.

Questions 3: Are the methods adequately described?

Response 3: Our manuscript has revised Figure 2 and added Sections 3.2.2 and 3.3.2 to describe the simulation methods.

Response to Reviewer 3 Comments

1. Summary

Thank you for your letter and insightful comments on our manuscript entitled “Ship Detection in Maritime Scenes Under Adverse Weather Conditions”. These comments not only greatly assist us in improving our paper, but also have important guiding significance for our subsequent research. We have studied the comments carefully and have made corrections to improve our paper. We appreciate for your warm work earnestly, and hope these corrections will meet with approval.

2. Questions for General Evaluation

Reviewer’s Evaluation

Response and Revisions

Does the introduction provide sufficient background and include all relevant references?

Can be improved

Are all the cited references relevant to the research?

Can be improved

Is the research design appropriate?

Can be improved

Are the methods adequately described?

Can be improved

Are the results clearly presented?

Can be improved

Are the conclusions supported by the results?

Can be improved

3. Point-by-point response to Comments and Suggestions for Authors

Comments 1: In figure 2, the output of the detection head is not very clearly visible. The bounding boxes are too small to be noticeable easily.

Response 1: Thank you for your suggestion. Your advice has been instrumental in improving the presentation of our ship detector framework. We have modified the ‘detection head’ in Figure 2 to include 'Classifier', 'Box Regressor', and 'Direction Classifier' modules.

Comments 2: In figure 4, the zoomed in portions have been captioned as ‘a’ and ‘b’ while the main figure has not been given any caption. Moreover, sub-figures ‘a’ and ‘b’ do not have very clearly marked boundaries. This makes it hard to understand the whole scenario.

Response 2: Thank you for your suggestions. The manuscript has been revised to include Figure 4, which adds visual displays of rainy, clear, and simulated rainy weather scenarios.

Comments 3: While discussing multi-scale 2dCNN in section 4.3, the abbreviation ‘RPN’  has been used without any explanation.

Response 3: Thank you for the reminder. We have explained "RPN" and cited relevant literature in the revised manuscript; detailed information is highlighted in yellow in the first sentence of Section 4.3.

Comments 4: Figure 6 describes the multi-scale 2d CNN module which is simply a variant of UNET with depth two. UNETs are routinely used by the practitioners in this field and as such cannot be claimed as a novelty.

Response 4: Thank you for your suggestions. The UNet network is primarily used in the image domain. In the 3D point cloud space, sparse convolution is utilized to extract 3D features that contain spatial information. These 3D features are then assembled into pseudo-images for feature extraction using a 2D convolution network. Previous works have only used 2D features for object classification, regression, and orientation classification. The dimensions of ships are much larger than those of vehicles, and this method loses information about the object's height. Instead, we use direct 3D features for object classification, regression, and orientation classification, addressing the issue of losing height information. This approach has not appeared in previous 3D object detection methods.

Comments 5: The respective sizes of the inputs and output in figures 5 and 6 have not been detailed.

Response 5: Thank you for your reminder. Detailed information can be found in the revised manuscript, specifically in Section 4.2 from the second to the third sentence of the first paragraph, and in Section 4.3 in the fourth sentence, both marked with a yellow background. Thank you once again for your reminder.

Comments 6: Table 4 caption misspells feature maps.

Response 6: Thank you for the reminder. We have made the correction in the revised version, and the details can be seen in the title of Table 4, highlighted in yellow.

Comments 7: More details of the lidar dataset need to be described. Is this dataset publicly available?

Response 7: The lidar dataset has been made public and originates from the article 'LiDAR Point Clouds Dataset of Ships in Maritime Environment,' with the DOI number 10.1109/JAS.2024.124275. The article has been accepted by the journal ‘IEEE/CAA Journal of Automatica Sinica’ but is not yet indexed.

4. Response to Comments on the Quality of English Language

Response: Thank you very much for your suggestions. We have meticulously checked the article's grammar multiple times to ensure its correctness and have made several revisions accordingly. We sought feedback from several native speakers to enhance the clarity, fluency, and professionalism of the article.

5.Questions for General Evaluation

Questions 1: Does the introduction provide sufficient background and include all relevant references?

Response 1: The revised manuscript has updated the introduction and added two references to provide a more comprehensive background. Specific details can be found in Section1, from the first to the third sentence of the third paragraph, which are highlighted with a yellow background.:

1. Application Research of Ship Overload Identification Algorithm Based on LiDAR Point Cloud

2. Preliminary Study for Motion Pose of Inshore Ships Based on Point Cloud: Estimation of Ship Berthing Angle.

Questions 2: Are all the cited references relevant to the research?

Response 2: We carefully reviewed each reference, removing those that did not fit the article and adding literature relevant to the field.

Questions 3: Is the research design appropriate?

Response 3: We revised the rainy day point cloud simulation process, adding Gaussian distribution filtering to the point clouds. More detailed information can be found in Section 3.3.1, from the fifth sentence to the end of the last paragraph, highlighted with a yellow background.

Questions 4: Are the methods adequately described?

Response 4: Our manuscript has revised Figure 2 and added Sections 3.2.2 and 3.3.2 to describe the simulation methods.

Questions 5: Are the results clearly presented?

Response 5: Our manuscript has revised Figures 3, Figures 4, and Table 1 to clearly display the simulation and detection results.

Questions 6: Are the conclusions supported by the results?

Questions 6: Our manuscript includes an addition to the third paragraph of Section 5.4 to describe the detection results.