A Two-Stage Semi-Supervised High Maneuvering Target Trajectory Data Classification Algorithm

Round 1

Reviewer 1 Report

1. The scheme presented in Figure 4 is very simplified, it should be corrected by presenting principled solutions

2. At Fig. 1, 2, 3 and others. at least minimal mathematical models should be presented, because without them the research novelty of the article is unclear.

3. Since there is no mathematical model of the results, it is not clear how the results presented in Table 1 were obtained. It is recommended to present a mathematical model of data processing in the article.

4. The given formulas (1) and (2) have no source reference, although they are known. It is recommended to clarify the sources of the formulas and describe their applicability to the problem analyzed in the article more precisely.

5. The graphs in Figure 8 should be arranged by giving the corresponding coordinate parameters.

6. The conclusions lack the clarity of the results achieved, i.e. what is achieved, what accuracies and so on. It is recommended to adjust the conclusions to reflect the achieved results more accurately.

7. The bibliography contains a number of sources that are 20 or more years old, although a number of works have been published in this area recently. It is recommended to update the literature list with new literature.

Author Response

Dear reviewer:

Thank you for your suggestion, I have made the modification according to your suggestion and marked the modified part in the paper.

The reply to your comments is as follows:

• According to the problem you pointed out that the scheme presented in Figure 4 is toosimplified,should be corrected by presenting principled solutions. We modified Figure 8 and introdced the working principle of the autoencoder and the loss calculation model
• Regarding the problem that At Fig. 1, 2, 3 and others. at least minimal mathematical models should be presented. We have added mathematical models such as loss function and accuracy calculation formula in the relevant part
• In response to the problem you pointed out that there is no mathematical model of the Table1. Considering that the corresponding mathematical models and indicators are given below, we adjust the table and analysis of this part to Section 4.3 of the experimental part, and introduce the corresponding experiments more detail.
• Regarding your suggestion, there is no source reference for the given metrics purity and divergence. Corresponding references are given and their related principles are described in more detail.
• In response to your suggestion that the graphs in Figure 8 should be arranged by giving the corresponding coordinate parameters.Considering that the trajectory in Figure 8 is normalized, its original longitude and latitude features are meaningless, so I added the original longitude and latitude trajectory image as an illustration.
• In response to your suggestion, the Conclusions section has been rewritten to more accurately reflect the results achieved.
• For your question about references, Part of the older literature is the introduction of the original classical algorithm, so it is retained. And 10 related references are added for explanation.

Thank you again for your valuable comments and hope to receive your reply as soon as possible, thank you.

Reviewer 2 Report

the abstract is well written.

The introduction should be supported with the previous literature where similar works had been conducted.

The problem is not well defined in the work. It should be highlighted.

The objectives are defined and however, the refinement is required to motivate the readers.

The problem definition is given in the paper and more details of the same different dimensions should be incorporated.

the Siamese network of theories are defined but requires better improvement for understanding of readers.

the experimental results are well presented with elaborate discussion on each figure and table.

more statistical results should be presented as a table.

the discussion still lacks in some figures and hence encouraged the authors to implement the same while submission of revision.

the citations are properly mentioned and should be in accordance with the journal style or format.

the conclusions are not well defined and not consistent with that projected in results section.

the English language should be edited at few instances of work.

overall, the current version of work needs revision before it is accepted for publication.

Author Response

Dear reviewer:

Thank you for your suggestion, I have made the modification according to your suggestion and marked the modified part in the paper.

The reply to your comments is as follows:

• According to the problem you pointed out that the introduction should be supported with the previous literature where similar works had been conducted.10 related references have been added to the paper for explanation.
• Regarding your suggestion that the problem is not well defined in the work and needs to be highlightedand detailed, the relevant part of the introduction has been revised.
• In response to the problem you pointed out that objectivesdefinition needs to be refined, a description to the article's objectives has been added to the end of the introduction.
• Regarding your suggestion that the Siamese network needs a more detailed introduction, We present a more detailed introduction to the principle of the Siamese network near Figure 2
• In response to the question you pointed out that the article needs more figures and tables, modified and added some figures and tables to illustrate the algorithm principle and experimental resultsFor your question about references, Added relevant references and revised format.
• In response to your suggestion, the Conclusions section has been rewritten to more accurately reflect the results achieved.
• The English in the text has been revised.

Thank you again for your valuable comments and hope to receive your reply as soon as possible, thank you.

Reviewer 3 Report

The experimental work presented in the Manuscript applsci-1918041, entitled " A Two-Stage Semi-Supervised High Maneuvering Target Trajectory Data Classification Algorithm " is interesting research with some promising results. The article reports a two-stage semi-supervised high maneuvering target trajectory data classification algorithm. Use Siamese network     to solve the problem of insufficient amount of labels. By pre-training the autoencoder and combin- ing it with the Siamese network, a two-stage joint training is formed, which enables the model to  have both clustering and classification capabilities, and solves the problem of missing categories, there are several shortcomings and modifications that should be included in order to enhance the final manuscript for the readers.

Abstract:

Please add conclusion of the abstract at the end?

Please write about the testing model comparing with training model in abstract?

Introduction

There was a lack of citations in introduction to present the value of this study. Please support the introduction by other study.

What is the novelty (originality) of the work? And what is new in your work that makes a difference in the body of knowledge? What has been done that goes beyond the existing research.

Please write the aim and objectives of this study at the end of introduction?

Line 46.   Reference [3], please add the name of the author for example,  Li et al . [3] as well as other such as  Reference [4], and  Reference [5].

Please, write the practical applications of your work in a separate section, before the conclusions and provide your good perspectives.

Figure 12. legend of x and y need to improve.

Conclusions

Please write about the limitations of this work in details in conclusion section.

Author Response

Dear reviewer:

Thank you for your suggestion, I have made the modification according to your suggestion and marked the modified part in the paper.

The reply to your comments is as follows:

• In response to your suggestion for the abstractsection, a summary is made at the end of the abstractand the difference between the compared models is made clear.
• Regarding the problem that there was a lack of citations in introduction, 10 related references have been added to the paper for explanation.
• Regarding your suggestion that the paper’s workis not well defined, The introduction of the article has been revised to better illustrate the work of this paper.
• You suggest that we write down the aim and objectivesof this study at the end of the introduction. We have modified as this request.
• Reference citation format has been revised.
• Before the conclusion, the practical application of the work is presented.
• Modified Figure 12 to add x, y coordinate labels.
• Regarding your suggestion that the limitations of this workneed to be introduced. The conclusion section has been revised to illustrate the limitations of this work.

Thank you again for your valuable comments and hope to receive your reply as soon as possible, thank you.

 

Reviewer 4 Report

General Comments:

The paper deals with data classification and presents a deep-learning based approach to classify high-maneuvering trajectories which may have some missing labels and missing category information. The proposed framework is a combination of Siamese network and an autoencoder, which can handle the above two problems jointly. Good results have been obtained versus the existing semi-supervised approach. The topic is an active area of research and the presented approach is interesting.

 

Specific Comments:

1.     Please define variables in Equation (1).

2.     Section 3:

a)    Sub-Section 3.1: The structure of the autoencoder should be explained in detail.

b)    Sub-Section 3.2: Please clarify whether there are any thresholds in the proposed Siamese structure.

c)     Sub-Section 3.3: The role of the KNN is unclear, so is the number of clusters K.

3.     As the work is dealing with high-maneuvering targets, please comment on real-time performance of the proposed network, e.g., by analyzing the time-complexity.

 

Language Usage: The paper is in need for a moderate language revision by a professional before it can be published.

Author Response

Dear reviewer:

Thank you for your suggestion, I have made the modification according to your suggestion and marked the modified part in the paper.

The reply to your comments is as follows:

• The variables in equation 1 are defined and introduced
• The structure of autoencoder in this paper is introduced in more detail
• As for your suggestion that the autoencoder needs to be introduced in more detail, we have added a table near Figure 5 to give a more detailed description of the automatic encoder structure used in this article.
• In Section 3.2, the use of the Siamesenetwork is introduced in more detail. In this paper, the Siamesenetwork is used to calculate the similarity between samples, and then the similarity matrix is used for clustering, so there is no need to set a threshold in this algorithm.
• The method of clustering using K-means algorithm and similarity matrix is introduced in detail. In practice, methods such as adaptive K-means algorithm should be used to determine the K value, but this algorithm in this paper is only used for comparative experiments, so the default is to know the K value for experiments.
• On the issue of time complexity, more complex network design will inevitably lead to greater computation. This paper points out this shortcomings of the algorithm in the final summary. But in fact, at present, the field mainly focuses on offline trajectory data processing, and real-time data processing is one of the current technical difficulties. Therefore, there is no requirement for the time complexity of network model training. So the problem of time complexity does not need to be considered for the time being, nor does this paper study time complexity.
• According to the problem you pointed out that the paper needs to be revised in English, I have carefully revised the paper and asked my teacher to proofread and check it.

Thank you again for your valuable comments and hope to receive your reply as soon as possible, thank you.

Reviewer 5 Report

General:

-        English language lacks readability and intelligibility in several places. The article needs thorough checking from this point of view, as it affects the understanding of the solutions in several places.

-        Please explain the terms in the presented formulae, especially at the beginning of the article.

-        The informational content of the work belongs to a very narrow niche of applications, namely the supervision of a target trajectory, in a context where insufficient data exists. Therefore, a more detailed description for the practical application of the proposed solution should be provided, especially because reading and understanding the article is difficult for unadvised readers. By this extension of the presentation, the area of possible interested readers may be increased.

 

-        The references list may be a good starting point for explaining practical applications for your solution: for example, does it help increasing vessel traffic safety (in case of AIS system operation), does it help estimate a flying object trajectory with a better accuracy, in security problems etc.   

The approach is based on a very rigid style in the fields of presenting algorithms, artificial intelligence-based solutions and data processing. Looking at the article in its entireness, the style employed is too abstract in my opinion and not so clearly presented, even if based on solid argumentation, and also supported by trustful experiments. I would recommend detailing some of the terms employed especially in Figures 1 and 2, for a better understanding of the rest of figures, and providing some less abstract information regarding the applicability of the provided solution.

- The aspect of Englis language to be worked out a little bit, it lacks intelligibility in some amount. While in general the paper is well written in English and grammar is good for any type of reader, I still recommend a thoroughly checking of English of the paper. The understanding of the meaning the authors desire to transmit is not easy, as stated before, especially for non-advised readers. Regarding the formulae presented in the article, some of them appear to need explanations of the employed terms.

 

Author Response

Dear reviewer:

Thank you for your suggestion, I have made the modification according to your suggestion and marked the modified part in the paper.

The reply to your comments is as follows:

• According to the problem you pointed out that the paper needs to be revised in English, I have carefully revised the paper and asked my teacher to proofread and check it.
• Your second and last suggestions respectively point out that formula terms in the text need to be introduced more specifically, and that relevant English grammar is difficult to understand. Therefore, we have introduced the terms in the formula introduced in this article in more detail.
• For your suggestions 3 and 4, we have revised the introduction and end of the article, introduced the application of this study in more detail, and explained the specific application of relevant references more fully. In fact, this paper is a part of the research project of our research group. Although it only aims at the special problems of insufficient label data sets and missing categories, this problem is widespread in the field of track classification and has high research value. And this problem exists not only in this field, but also in other fields where neural network models are used for classification. We believe that this algorithm can be used as a technical solution to such problems, not just for specific applications. So we believe that this program can arouse the interest of relevant readers.
• In view of the problem that you pointed out that the explanation in the article is too abstract, the relevant experiments and terms, especially the contents related to Figures 1 and 2, are introduced in more detail.

Thank you again for your valuable comments and hope to receive your reply as soon as possible, thank you.

Round 2

Reviewer 1 Report

It is recommended to clarify the sources of the formulas and describe their applicability to the problem analyzed in the article more precisely.

Author Response

Dear reviewer:

Thank you for your suggestion, I have made the modification according to your suggestion and marked the modified part in the paper.

In response to your suggestion to clarify the sources of the formulas and describe their applicability to the problem analyzed in the article, I have revised the formula introduction section of the article to introduce the rationale for these formulas and to highlight the reasons for their selection. 

Thank you again for your valuable comments and hope to receive your reply as soon as possible, thank you.

Reviewer 3 Report

Please put the practical application before the conclusion of the manuscript. 

Author Response

Dear reviewer:

Thank you for your suggestion, I have made the modification according to your suggestion and marked the modified part in the paper.

In response to your suggestion to put the practical application before the conclusion of the manuscript, I present an introduction to the role of the algorithm in practical application in section 5.1 of the conclusion section.

Thank you again for your valuable comments and hope to receive your reply as soon as possible, thank you.

Reviewer 4 Report

The Authors have revised the manuscript and added sufficient details to clarify the Reviewer’s concerns. The current version deserves publication.

Reviewer 5 Report

The recommendations for improving the quality of the article have been met.