Multiple Mobile Target Detection and Tracking in Small Active Sonar Array

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The introduction provides sufficient background  and contextualization for using the method in underwater scenarios for object tracking and biodiversity monitoring and other marine apps. Additional references could strengthen the discussion of novel developments related to adaptive beamforming techniques in underwater context (Alford, S., et al. (2022). “Advancements in Underwater Beamforming for Mobile Platforms.” Ocean Engineering, 112040. https://doi.org/10.1016/j.oceaneng.2022.112040). The methods are adequately described. The use of a debiased converted measurement using a Kalman filter, CFAR detection, and blob tracking are clearly articulated, also a diagrams assist in understanding the experimental setup. The results are clearly presented both for simulations and real sea trials with apparent consistent results. A doubt about how the level of the SCR was defined, estimated or  measured is required for clarification. The conclusions logically follow the achieved results, showing the improved track continuity and reduced false tracks that are effectively demonstrated through simulation and real-world data.

I consider it is important to discuss potential limitations of the method (for instance, handling overlapping targets with similar velocities, the cross section of the object and the incidence angles), or consider adding reproducibility statements or code/dataset availability

Comments for author File: Comments.zip

Author Response

Comment 1: Additional references could strengthen the discussion of novel developments related to adaptive beamforming techniques in underwater context (Alford, S., et al. (2022). “Advancements in Underwater Beamforming for Mobile Platforms.” Ocean Engineering, 112040. https://doi.org/10.1016/j.oceaneng.2022.112040). 

Response 1: Thank you for your suggestion. We were not able to locate the suggested paper (the DOI points to another paper, and google scholar cannot find the title). We did however added a few more references now cited as  [8], [15], [16], [17], [18] and [19].

 

Comment 2: A doubt about how the level of the SCR was defined, estimated or  measured is required for clarification.

Response 2: The level of SCR was defined at the output of the matched filter matrix before blob creation. The signal is the range and angle of the target as known by GPS, the noise is the average of all other samples. This has been clarified in the text at the first time the SCR appears.

 

Comment 3: I consider it is important to discuss potential limitations of the method (for instance, handling overlapping targets with similar velocities, the cross section of the object and the incidence angles), or consider adding reproducibility statements or code/dataset availability

Response 3: In the revised conclusion section, we have added text on the limitation of our method. These exists when the targets cross each other or rapidly change their speed. Unfortunately, due to IP limitations of our institutes, we cannot release the implementation code.

Comments highlighted in the paper has all been addressed:

1. Repetition of the acronym declaration: constant false alarm.
2. Defenision of the acronym for GPS.
3. Definition of SCR.
4.  Fix of NCV acronym.
5. Text below (20): Fixed.
6. Answer to comment above Fig.4 is in Response 2 above.
7. Regarding the same figures as archiveX paper, we understand that pre-publishing at archiveX is allowed by MDPI.
8. Deadlink in reference [23]: the paper has been published and the refernece updated.

Reviewer 2 Report

Comments and Suggestions for Authors

1- It is stated that the proposed system performs detection and tracking in batch form to mitigate occlusion and multipath effects.  Is the association done across multiple frames, or only one frame at a time?

2- Abstract: Clarify what “2D time-space matrix” means—does it refer to angle vs. distance?

3- The use of GPS-tagged floats is creative, but the maximum localization error of 24 m is very high. For a more precise comparison, consider: Using acoustic tags with time-difference-of-arrival (TDOA) references.
Comparing estimated velocity vectors against high-resolution tracks to validate dynamic behavior.

4- Include a comparison with other dynamical models (e.g., IMM-based tracking) to show that the CV model is sufficient.

5- The choice of 4-connectivity in blob detection should be justified. Would 8-connectivity improve robustness to fragmented detections?

6- Some important works in underwater tracking and sonar image processing are missing. Including foundational references such as [Blackman & Popoli] or more recent probabilistic data association models would strengthen the context.

7- More detailed quantitative comparisons (e.g., RMSE, ID switch count, F1 score) should be added alongside track continuity.

Author Response

Comment 1: It is stated that the proposed system performs detection and tracking in batch form to mitigate occlusion and multipath effects.  Is the association done across multiple frames, or only one frame at a time?

Response 1: Indeed, the target association is done across several frames and the decision is made at the end of the batch. We have clarified this in the revised introduction.

 

Comment 2:  Abstract: Clarify what “2D time-space matrix” means—does it refer to angle vs. distance?

Response 2: This refers to time vs. distance. Clarified in the revised abstract.

 

Comment 3: The use of GPS-tagged floats is creative, but the maximum localization error of 24 m is very high. For a more precise comparison, consider: Using acoustic tags with time-difference-of-arrival (TDOA) references. Comparing estimated velocity vectors against high-resolution tracks to validate dynamic behavior.

Response 3: Using acoustic tags is indeed a good option. However, our experiance with such tags (see for example our technical paper in (Talmon Alexandri, Eyal Miller, Ehud Spanier, and Roee Diamant, ``Tracking the Slipper Lobster using Acoustic Tagging: Testbed Description", IEEE Journal of Ocean Engineering, vol. 45, no. 2, pp. 577-585, 2020) show that the tags’ signal generation rate is on the order of tens of seconds at best. At this rate, the updates for the fish location will be too slow to provide meaningfull ground truth. Instead, we require a trajectory update rate of less than a sec, which is the reason we picked the GPS floats. This discussion as been added to the Results section.

 

Comment 4: Include a comparison with other dynamical models (e.g., IMM-based tracking) to show that the CV model is sufficient.

Response 4: We benchmarked our filtering scheme with two other schemes: The “Lo” (reference [31]) and the “Karoui” (reference [32]). We realize that their are a large number of other benchmarks that can be used, but we picked these two benchmarks beacuse they have been spesifically tested for underwater target trecking. For the proposed IMM filter, we argue that the comparison will not be fair. First, the IMM filter was made for a sinlge target tracking while ours is set for multiple target tracking. Second, the IMM requires (multiple) dynamic models while ours does not assume one. Third, the IMM is a state-space tracking algorithms, while ours is a batch filter, more similar to the track-before-detect approach. For these reasons, if that is OK with the reviewer, we perfer to maintain our benchmarking with the above two approaches.  

 

Comment 5: The choice of 4-connectivity in blob detection should be justified. Would 8-connectivity improve robustness to fragmented detections?

Response 5: Indeed, higher dimensional connectivity may improve the estimation of the blob's center of mass. However, the change is expected to be minor (order of a few cm), which is also smoothed by the filtering operation. The solution using the 4-connectivity was found robust enough to detect connected blobs.

 

Comment 6: Some important works in underwater tracking and sonar image processing are missing. Including foundational references such as [Blackman & Popoli] or more recent probabilistic data association models would strengthen the context.

Response 6: In the revised manuscript, we have added citations of such works in citations [8], [15], [16], [17], [18] and [19]. These are embedded in our list of 50 citations, all relevant to our application.

 

Comment 7: More detailed quantitative comparisons (e.g., RMSE, ID switch count, F1 score) should be added alongside track continuity.

Response 7: The reason we choose to focus on track contineuity is because it encompasses both the location error (e.g., RMSE) and the tracking error (e.g., ID switch). Our definision of the track contineuity considers if the location of the target is indeed where the real target is, otherwise the score decreases. On top of this analysis, we provided some example results of estimated tracks for both simulations and sea trials. But more importently, the reason we measure and compare performance by track contineuity only is that our main goal is to find the valid targets and to accurately characterize their dynamics rather than providing true location estimate. We argue that this is best captured by our metric. Thus, if that is OK with the reviewer, we prefer to maintain our analysis for the track contineuity only. We have added a clarification for this in Section 6.1. 

Reviewer 3 Report

Comments and Suggestions for Authors

For the multiple mobile target detection and tracking in small active sonar array, the authors present an algorithm for detecting and tracking mobile underwater objects using reflections from active acoustic emission of broadband signals received by a rigid hydrophone array. The proposed algorithm can overcome the problem of high false alarm rate by applying a tracking approach to the sequence of received reflections. The proposed algorithm has a certain degree of innovation, but there are the following issues that urgently need improvement.

1. It is noted that the authors’ manuscript needs careful editing by someone with expertise in technical English editing paying particular attention to English grammar, spelling, and sentence structure so that the goals and results of the study are clear to the reader. For example, “The detection and tracking of underwater objects is required for applications such as the monitoring of marine fauna [1], the identification of schools of fish in commercial ventures [2] and the detection of prey of sperm whales (Physeter macrocephalus) for behavioral analysis [3].”, “a high ratio between the duration of the track and the lifetime of the target”, etc. There are too many similar issues, so I will not list them one by one here. At the same time, there are many long sentences. It is suggested to change them into sentences that are easy to understand.
2. In small active sonar array, “target” should be used, not “object”. Suggest checking the entire paper and correcting similar issues.
3. When the abbreviation first appears, the full name should be given, such as “CETI”. It is recommended to check the entire paper and modify similar issues.
4. The paper has many non-standard expressions. For example, “Pictures of our glider and floater platforms can be found in Fig 1a and 1b respectively”. It is recommended to modify as “Pictures of our glider and floater platforms can be found in Fig 1a and 1b, respectively”.
5. The authors only used the simplest Kalman filter suitable for linear Gaussian filtering. Considering complex nonlinear and non Gaussian situations, how should the authors handle it?
6. Considering the engineering implementation of the proposed algorithm, the authors should provide a specific process and relevant pseudocode.

Author Response

Comment 1: It is noted that the authors’ manuscript needs careful editing by someone with expertise in technical English editing paying particular attention to English grammar, spelling, and sentence structure so that the goals and results of the study are clear to the reader. For example, “The detection and tracking of underwater objects is required for applications such as the monitoring of marine fauna [1], the identification of schools of fish in commercial ventures [2] and the detection of prey of sperm whales (Physeter macrocephalus) for behavioral analysis [3].”, “a high ratio between the duration of the track and the lifetime of the target”, etc. There are too many similar issues, so I will not list them one by one here. At the same time, there are many long sentences. It is suggested to change them into sentences that are easy to understand.

Response 1: We have streamedline our manuscript and considerably edited the text. 

 

Comment 2: In small active sonar array, “target” should be used, not “object”. Suggest checking the entire paper and correcting similar issues.

Response 2: We have replaced all “object” instances with “target” throughout.

 

Comment 3: When the abbreviation first appears, the full name should be given, such as “CETI”. It is recommended to check the entire paper and modify similar issues.

Response 3: We have streamline our manuscript and corrected all such matters. 

 

Comment 4: The paper has many non-standard expressions. For example, “Pictures of our glider and floater platforms can be found in Fig 1a and 1b respectively”. It is recommended to modify as “Pictures of our glider and floater platforms can be found in Fig 1a and 1b, respectively”.

Response 4: We have corrected this quoted text, and have streamline the manuscript for other similar issues.

 

Comment 5: The authors only used the simplest Kalman filter suitable for linear Gaussian filtering. Considering complex nonlinear and non Gaussian situations, how should the authors handle it?

Response 5: We clarify that we use the debiased Kalman filter and not the simple one. The debiased Kalman filter converts polar to Cartesian measurement as a fast and efficient method to handle a non linear measurement model. This is explained at the end of section 4.c.

 

Comment 6: Considering the engineering implementation of the proposed algorithm, the authors should provide a specific process and relevant pseudocode.

Response 6: We would like to refer the reviewer to the block diagram in Fig. 2, which describes the prcoess of the solution. This block diagram is exactly the steps of the solution, and thus, if that is OK with the reviewer, we wish to avoid a pseodocode which will assentially be the same as the floaw chart just in a text form.