A Fast Dynamic Dim Target Tracking Approach for UAVs Using Improved Lightweight Siamese Networks

Round 1

Reviewer 1 Report

The article presents propose an improved lightweight siamese 16 network (ILSN) with an optimized design about the feature extraction network and the similarity measurement

 

Presented problem and obtained results are interesting. However, the article has some weaknesses and points worth explaining.

 

 

 

- Authors should highlight their contribution in the introduction

 

- Authors present results obtained by simulations and from UAV experimental system

 

- The data collection process in a real experiment should be described in more detail

 

- The description of the presented results could be extended (I.e. figure 9)

 

- Conclusions should be extended ( I.e A comparison of the simulation and experimental results should be made

 

 

- line 323 Table I ( maybe Table 1)

Author Response

Dear Reviewers，

 

Thank you for your efficient and professional work (manuscript ID: applsci-2014629).

We have completed the revision of the manuscript and responded to your comments one by one. All the modifications are highlighted in the manuscript.

Thanks again for your valuable comments. We hope our work will meet the requirements of your journal!

 

Best regards!

Lianpeng Li, Xu zhao et al.

Reviewer:1

Comment to the Author:

The article presents propose an improved lightweight siamese network (ILSN) with an optimized design about the feature extraction network and the similarity measurement. Presented problem and obtained results are interesting. However, the article has some weaknesses and points worth explaining.

Comment 1:

Authors should highlight their contribution in the introduction.

Response 1:

 Thank you for this comment. As you said, The presentation of the contribution of this paper in the introduction is not significant enough. For this, we have reorganized the logic of the introduction to highlight the necessity of our research work. Moreover, the description of the contribution of this paper is optimized at the end of the introduction, emphasizing the core work of the proposed method, model and experimental system.

For example, the main contribution of this paper is re-summarized as follows:

1) We propose an improved lightweight siamese network (ILSN) with an optimized design for the feature extraction network and the similarity measurement. ILSN has obvious advantages in tracking accuracy and tracking speed, which has been demoted by simulations and system test.

2) We optimize the feature network extraction capability by designing a position-wise attention module and localization information fusion, which in turn improves the accuracy and real-time performance of ISLN target tracking;

3) Finally, we establish a simulation environment and a UAV-based experimental platform. By comparing with comparison algorithms such as SiamRPN, DaSiamRPN, and SlamRPN++, we verify the fallibility and feasibility of the ILSN.

Thank you for your valuable comments, we think these changes will make the contribution more prominent. All of our changes are highlighted in the manuscript on pages 1 and 2.

Comment 2:

Authors present results obtained by simulations and from the UAV experimental system.

Response 2:

Thank you for your professional review. We conduct algorithm simulation experiments to verify the effectiveness and feasibility of ILSN by using KCF, MCNN, SiamRPN, DaSiamRPN, and SlamRPN++ as comparison algorithms. Through simulation experiments, the efficiency of the algorithm is initially demonstrated. To further demonstrate the effectiveness of the algorithm in a physical system, we built the experimental platform built based on DJI Mini 3 pro.

Our representation in this section was indeed not clear enough. The improvements are as follows:

• From lines 340 to 345, we redescribe the purpose of the simulation system and the physical experiment platform;
• In line 431, we highlight the performance advantages of the proposed method in a simulation system;
• In line 481, We give the main purpose of establishing the experimental platform, which is to verify the feasibility of ILSN.

Comment 3:

The data collection process in a real experiment should be described in more detail.

Response 3:

We do describe the experimental system data acquisition process inadequately. For this, we have given scene selection, UAS pre-configuration, monitor control, data transfer process and data quantities from 495 to 504. 

The modifications in the manuscript are：

We chose eight different temporal, spatial, and altitude data acquisition scenarios, including city, mountain, street and lake, in which we first set up the server data acquisition environment and set up the monitor and UAV communication channels. The UAV system dynamically captures low-altitude images based on the ultra-high-definition binocular vision camera, which is controlled by the monitor. The monitor allows us to dynamically adjust the speed and height of the drone to increase the variety of samples collected. The collected data has been transmitted to the server and monitor through the image transmission module. The server stores and processes the data according to the obtained images completes the dynamic target recognition and transmits the data to the flight control system to complete the target tracking.

Then, we obtained 1000 sets of image data for each of the 8 scenarios through post-processing. An example of the collected experimental data is shown in Figure 7, which shows the target recognition and tracking images of 8 scenes, covering various scenes such as night, mountain, lake, and high speed.

Comment 4:

The description of the presented results could be extended (I.e. figure 9).

Response 4:

As you said, we have not described some of the results adequately. We have revised the article in the light of your comments.

For example:

• As for Figure 9, we have added the analysis of the graphic content from lines 533 to 536.
• As for Figure 8, We have added a description of the details of the image content from lines 522 to 523.

All changes have been highlighted in this manuscript.

Comment 5:

Conclusions should be extended ( I.e A comparison of the simulation and experimental results should be made.

Response 5:

According to your comments, we made the following changes:

• We have expanded the experimental conclusions section to analyze the performance differences between the simulated and physical experiments and to obtain comparative results.
• We have optimized the conclusions by adding a comparison of the simulation and experimental results, and also by pointing out the shortcomings of this study.

We are grateful for this comment, which is a significant improvement in the quality of our articles. All the changes are highlighted on page 16.

Comment 6:

line 323 Table I ( maybe Table 1)

Response 6:

Thank you for your careful review, we made some simple mistakes. For these issues, we have implemented the following corrections.

1) “Table I” has been replaced by Table 1.

2) Accuracy of expression，i.e. “week and small target” has been replaced by “dim target” in line 343.

3) Grammatical errors, i.e. “a series of researches” has been replaced by “a series of research” in line 93.

Thank you for your careful review and improvement of the quality of the, we appreciate it.

 

 

 

 

Author Response File: Author Response.pdf

Reviewer 2 Report

Navigation in a dynamic underwater environment is a very significant but difficult problem with a wide range of applications. Authors attempt to solve the problem using the now popular Sparrow Search Algorithm (SSA). It is well known that while SSA provides fast convergence, it also suffers from issues such as premature convergence to local optima. The authors solve to alleviate this issue by adding improvements and evaluating the Improved SSA. The methodology is fine, in general, however, there are several critical points that have to be addressed.

The paper suffers from minor (and sometimes major) issues with the formal description of the problem and the algorithm. Authors do not shy away from formulating the underlying optimization problem. However, answers to many emerging questions remain unclear.

- What are the decision variables?

- It is arguable whether the quality coefficient as defined in (1) has a reasonable interpretation. Here added are two position-dependent values and an energy-related value. What is the interpretation of the coefficient as a whole? How should one pick the weight coefficients (omega) or the threat factor?

- The authors claim that they include the dynamics of the environment such as time-varying current characteristics. However, in the model (equation 7) this parameter is given as a constant - there is no dependency on time.

- The experiments contained within the paper are very limited. They form no ground for the statement that the Improved SSA is better in general. They are an illustration that for some cases, the Improved SSA can perform well.

- The execution time of algorithms tested in the experiments is limited to the number of iterations. However, what is the real execution time? It is very well possible, that ISSA is actually slower convergence-wise when real time is considered (slow search speed is a flaw of SSA after all).

Finally, what follows are some less pressing but sill significant issues with the paper:
- How significant is the 90deg angle constraint (3)? Parameter "alpha_max" is not used in the cost function (1) so is it irrelevant?

- In equation (6) the coefficient "j" is unbounded.

- Is "modulo value of the path point "g_i" to "g_{i+1}" (page 3) simply the Euclidean distance?

- There is no need to introduce the constant "a" if it is set to one anyway. The same for "ST".

- What is the parameter "d"?

- Is variable "Q" (equations 8 and 10) generated once, or does it change with iterations of the algorithm. If it is the latter, this should be reflected in the notation (iteration number).

- In problem formulation the fitness function is denoted as "F". However, in algorithm description then "f(X_{i, j})" is used.

- In equation (14) and the subsequent description there are two terms used: "cauhy", "Gauhy". Are they spelling mistakes?

- It is said that a spline is used to smooth out the trajectory. But this can make constraints infeasible (such as constraint 4).

Author Response

Dear Reviewers，

 

Thank you for your efficient and professional work (manuscript ID: applsci-2014629).

Thank you for your valuable comment. There seems to be a mistake between the content of your comment and this article, which focuses on "A Fast Dynamic Dim Target Tracking Approach for UAVs Us-ing Improved Lightweight Siamese Networks". How to efficiently and accurately identify dynamic dim targets in complex contexts poses a challenge. We propose an improved lightweight siamese network (ILSN) with an optimized design about the feature extraction network and the similarity measurement.

We are very grateful for your hard work and we admire the rigorous wording and analysis in your comments. We have included below questions and answers from other reviewers to this paper for your review.

 

Best regards!

Lianpeng Li, Xu zhao et al.

 

Comment to the Author:

Navigation in a dynamic underwater environment is a very significant but difficult problem with a wide range of applications. Authors attempt to solve the problem using the now popular Sparrow Search Algorithm (SSA). It is well known that while SSA provides fast convergence, it also suffers from issues such as premature convergence to local optima. The authors solve to alleviate this issue by adding improvements and evaluating the Improved SSA. The methodology is fine, in general, however, there are several critical points that have to be addressed.

The paper suffers from minor (and sometimes major) issues with the formal description of the problem and the algorithm. Authors do not shy away from formulating the underlying optimization problem. However, answers to many emerging questions remain unclear.

 

Comment to the Author:

The article presents propose an improved lightweight siamese network (ILSN) with an optimized design about the feature extraction network and the similarity measurement. Presented problem and obtained results are interesting. However, the article has some weaknesses and points worth explaining.

Comment 1:

Authors should highlight their contribution in the introduction.

Response 1:

 Thank you for this comment. As you said, The presentation of the contribution of this paper in the introduction is not significant enough. For this, we have reorganized the logic of the introduction to highlight the necessity of our research work. Moreover, the description of the contribution of this paper is optimized at the end of the introduction, emphasizing the core work of the proposed method, model and experimental system.

For example, the main contribution of this paper is re-summarized as follows:

1) We propose an improved lightweight siamese network (ILSN) with an optimized design for the feature extraction network and the similarity measurement. ILSN has obvious advantages in tracking accuracy and tracking speed, which has been demoted by simulations and system test.

2) We optimize the feature network extraction capability by designing a position-wise attention module and localization information fusion, which in turn improves the accuracy and real-time performance of ISLN target tracking;

3) Finally, we establish a simulation environment and a UAV-based experimental platform. By comparing with comparison algorithms such as SiamRPN, DaSiamRPN, and SlamRPN++, we verify the fallibility and feasibility of the ILSN.

Thank you for your valuable comments, we think these changes will make the contribution more prominent. All of our changes are highlighted in the manuscript on pages 1 and 2.

Comment 2:

Authors present results obtained by simulations and from the UAV experimental system.

Response 2:

Thank you for your professional review. We conduct algorithm simulation experiments to verify the effectiveness and feasibility of ILSN by using KCF, MCNN, SiamRPN, DaSiamRPN, and SlamRPN++ as comparison algorithms. Through simulation experiments, the efficiency of the algorithm is initially demonstrated. To further demonstrate the effectiveness of the algorithm in a physical system, we built the experimental platform built based on DJI Mini 3 pro.

Our representation in this section was indeed not clear enough. The improvements are as follows:

• From lines 340 to 345, we redescribe the purpose of the simulation system and the physical experiment platform;
• In line 431, we highlight the performance advantages of the proposed method in a simulation system;
• In line 481, We give the main purpose of establishing the experimental platform, which is to verify the feasibility of ILSN.

Comment 3:

The data collection process in a real experiment should be described in more detail.

Response 3:

We do describe the experimental system data acquisition process inadequately. For this, we have given scene selection, UAS pre-configuration, monitor control, data transfer process and data quantities from 495 to 504. 

The modifications in the manuscript are：

We chose eight different temporal, spatial, and altitude data acquisition scenarios, including city, mountain, street and lake, in which we first set up the server data acquisition environment and set up the monitor and UAV communication channels. The UAV system dynamically captures low-altitude images based on the ultra-high-definition binocular vision camera, which is controlled by the monitor. The monitor allows us to dynamically adjust the speed and height of the drone to increase the variety of samples collected. The collected data has been transmitted to the server and monitor through the image transmission module. The server stores and processes the data according to the obtained images completes the dynamic target recognition and transmits the data to the flight control system to complete the target tracking.

Then, we obtained 1000 sets of image data for each of the 8 scenarios through post-processing. An example of the collected experimental data is shown in Figure 7, which shows the target recognition and tracking images of 8 scenes, covering various scenes such as night, mountain, lake, and high speed.

Comment 4:

The description of the presented results could be extended (I.e. figure 9).

Response 4:

As you said, we have not described some of the results adequately. We have revised the article in the light of your comments.

For example:

• As for Figure 9, we have added the analysis of the graphic content from lines 533 to 536.
• As for Figure 8, We have added a description of the details of the image content from lines 522 to 523.

All changes have been highlighted in this manuscript.

Comment 5:

Conclusions should be extended ( I.e A comparison of the simulation and experimental results should be made.

Response 5:

According to your comments, we made the following changes:

• We have expanded the experimental conclusions section to analyze the performance differences between the simulated and physical experiments and to obtain comparative results.
• We have optimized the conclusions by adding a comparison of the simulation and experimental results, and also by pointing out the shortcomings of this study.

We are grateful for this comment, which is a significant improvement in the quality of our articles. All the changes are highlighted on page 16.

Comment 6:

line 323 Table I ( maybe Table 1)

Response 6:

Thank you for your careful review, we made some simple mistakes. For these issues, we have implemented the following corrections.

1) “Table I” has been replaced by Table 1.

2) Accuracy of expression，i.e. “week and small target” has been replaced by “dim target” in line 343.

3) Grammatical errors, i.e. “a series of researches” has been replaced by “a series of research” in line 93.

Author Response File: Author Response.pdf