Automated VIIRS Boat Detection Based on Machine Learning and Its Application to Monitoring Fisheries in the East China Sea

Round 1

Reviewer 1 Report

The present study proposes a novel methodology based on machine learning to detect fishing vessels from VIIRS nighttime light imagery. Previous algorithms used for extracting light luring fishing vessels from VIIRS/DNB data relied on a series of complex thresholds, which were challenging to determine, particularly in periods of cloudy weather and moonlight. The machine learning method presented here successfully resolves this issue and produces comparable results. In this way, the machine learning algorithm for vessel detection applied in this study represents a major advancement in the monitoring of fishing vessels with nighttime light remote sensing. I believe that this manuscript should be published in Remote Sensing, with some revisions.

Particular comments:

1. The data and processes of the two traditional methods (EOG and FRA) have not been fully elucidated or adequately compared. Elvidge's VBD paper outlined a system for detecting boats in moonless settings, yet this setup can generate a large number of false detections when moonlight is available. In order to counteract this, EOG proposed a self-adjusting SMI threshold to reduce the false detections. Both EOG and FRA's algorithms require a number of thresholds to filter the vessel detection, and higher thresholds during conditions such as those with moon and cloud can lead to the removal of legitimate fishing vessels. A more comprehensive description of the data and processes of these two algorithms is needed in order to accurately assess and compare their efficacy.

2. A total of 41 days of radar data were utilized to assess the machine learning algorithm, with only one day explicitly mentioned in the manuscript, indicating a precision score of 0.92. These assessment data are critical information, which should be thoroughly detailed and elucidated.

3. The use of nighttime imagery to monitor fishing vessels in near-real time comes with certain caveats and limitations. To better understand the disadvantages of existing algorithms for extracting fishing vessels from VIIRS/DNB, the author could provide daily counts of vessels VS moon-phase and cloud conditions. Deep learning method may also provide better performance for fishing monitoring in the future work with more ground truth data.

4. In the introduction section, more references that make use of VIIRS/DNB to examine the light-attracting fishing vessels should be included, thereby providing a more comprehensive overview of the area and the primary fishing gears that have been studied.

For example:

Li J, Qiu Y, Cai Y, et al. Trend in fishing activity in the open South China Sea estimated from remote sensing of the lights used at night by fishing vessels[J]. ICES Journal of Marine Science, 2022, 79(1): 230-241.

Exeter O M, Htut T, Kerry C R, et al. Shining light on data-poor coastal fisheries[J]. Frontiers in Marine Science, 2021, 7: 625766.

5. Line 238: Only should be only

6. Line 537: his should be This

The language is generally fluent and readable, but there are some minor grammatical and spelling errors. Paying more attention to them would be even better.

Author Response

Dear Reviewer
We would like to express our heartfelt appreciation to the reviewer for your valuable feedback on our manuscript. Your expertise and thoughtful comments have immensely contributed to improving the quality and rigor of our research. We are truly grateful for your time, dedication, and commitment to maintaining the standards of scientific excellence. Your suggestions and critiques have played a crucial role in enhancing the overall impact of our work. Thank you for your invaluable contributions to our study.

Sincerely,
Masaki Tsuda

Author Response File: Author Response.docx

Reviewer 2 Report

The paper develops an algorithm for vessel detection using SAR images, particularly focusing on addressing VIIRS data imbalance. A two-step procedure is used to address this challenge. First a random forest classier is trained using prior works. The trained model is then used for vessel detection.

Experimental results on real-world data are provided to demonstrate that the proposed algorithm is effective and leads to performances comparable with VIIRS boat detection algorithms.   The proposed architecture is novel and the experiments are somewhat convincing. I have the following comments to be addressed for the next round of reviews:   1.  An idea to improve vessel detection in sea images using SAR images is to benefit from transfer learning. Consider the following works:   a. Zhang, H., Zhang, X., Meng, G., Guo, C. and Jiang, Z., 2022. Few-Shot Multi-Class Ship Detection in Remote Sensing Images Using Attention Feature Map and Multi-Relation Detector. Remote Sensing, 14(12), p.2790.   b. Lu, C. and Li, W., 2018. Ship classification in high-resolution SAR images via transfer learning with small training dataset. Sensors, 19(1), p.63.   c. Rostami, M., Kolouri, S., Eaton, E. and Kim, K., 2019. Deep transfer learning for few-shot SAR image classification. Remote Sensing, 11(11), p.1374.   The above works should be discussed in the introduction to give the reader a broader perspective on ship detection.   2. Section 2.1 is very brief and challenging for a reader that is not familiar with the architecture. Please add more description in the text to explain the architecture.   3. In the results, please run the experiments several times and report both the average and the standard deviation to make comparisons statistically meaningful.     4. Please add 2-3 existing baselines for comparison to demonstrate how competitive the results are.   5. Could you release the code and data on a public domain such as GitHub to make reproducibility easy?        

The paper can be followed well.

Author Response

Dear Reviewer
We would like to express our heartfelt appreciation to the reviewer for your valuable feedback on our manuscript. Your expertise and thoughtful comments have immensely contributed to improving the quality and rigor of our research. We are truly grateful for your time, dedication, and commitment to maintaining the standards of scientific excellence. Your suggestions and critiques have played a crucial role in enhancing the overall impact of our work. Thank you for your invaluable contributions to our study.

Sincerely,
Masaki Tsuda

Author Response File: Author Response.docx

Reviewer 3 Report

Comments for author File: Comments.pdf

Author Response

Dear Reviewer
We would like to express our heartfelt appreciation to the reviewer for your valuable feedback on our manuscript. Your expertise and thoughtful comments have immensely contributed to improving the quality and rigor of our research. We are truly grateful for your time, dedication, and commitment to maintaining the standards of scientific excellence. Your suggestions and critiques have played a crucial role in enhancing the overall impact of our work. Thank you for your invaluable contributions to our study.

Sincerely,
Masaki Tsuda

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

I am convinced by the authors' responses.