Time-Series Correlation Optimization for Forest Fire Tracking
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsCritical review on a paper entitled “Time series correlation optimization for forest fire tracking” submitted by Dongmei Yang and others to Forests
The manuscript under consideration is dedicated to important problem of forest fire observation and tracking using time series analysis and machine learning. However, quality of presentation of the research results is questionable.
The target audience of Forests journal are researcher and engineers of forestry and forest services. Clearly, they are not computer vision or machine learning engineers. So, the importance of the conducted study and implementation of the findings have to be clear. Started the abstract section context and possible implementation of the proposed solutions should be clearly explained. Are these observation towers or UAV/orbital imaging? Can these software solutions work autonomously or require mainframe/cloud computational resources? In which way existing solutions are flawed, so it worth to invest into the newer research in that direction?
The commonly used explanation pattern of following from the general concepts to discreet facts could REALLY help authors to bring your ideas to readers. So far, authors bring orphaned and somehow isolated statements which is hard to outline and follow. Please explain from broader to narrower terms, do not make readers to guess! They even write on convolution neural networks without mentioning convolution neural networks!
Also, the specificity of the audience needs more basic explanation of the machine learning models architecture. Legends of the model schemes are missing, numerous and dispersed over the text abbreviations greatly impedes the reading. Crucial references on the sources are also needed. Vague and wordy explanations may repulse the audience. Also, structure of the paper may need improvements including clarification and shortening. So far, the structure of the paper is rather frustrating. For instance, obviously labor under consideration it is not a review of the research field, it is a dedicated research. But authors commit the ENTIRE section 2 to “Related Works”. In case of dedicated research it is normal to share references to predecessors between “Introduction” and “Results and discussion” sections, but authors commit the ENTIRE section 2 to “Related Works”. “Materials and Methods” sections. I would recommend to change the paper structure making it robust data-driven research.
Besides of that base discrepancies, existing typos, errors that show lack of proofreading worsen impression of that written labor. Some of them are enlisted below:
Line 2. “UAV based suppression” presumes direct action onto fire. However is was not mentioned anywhere else.
L6. To what area of machine learning/computaion/modeling/computer vision that “AO-OCSORT framework” is relevant? What is the context?
L32. Here and everywhere IN THE TEXT. The reference numbers were substituted by the question marks (“[?]”), so it is not possible to track references. How it comes that is was not noticed?
L48. Distance from the observation point, spatial, time and spectral resolution could facilitate the understanding.
L59-78. That paragraph could fit Material and Methods section better.
L62. What the “Mamba architecture” is? No reference is provided and no context is given.
Section 2.3 Which value the section 2.3 gives to the text?
Figures 1; 3. Please, provide the abbreviations in the description of the image of its legend. Also, provide meaning of the different colors and shapes of the blocks.
L189; 376; 393; 402; 410; 418; 432; 441; 451;463;468;476;. Orphaned heading? Where is its number?
Figure 3. “Motion detection architecture”. Architecture of what?
Section 3.2. So far, reading still undisclosed the structure of data inputs, software used for machine learning model development, API for communication between modules and hardware implementation. According the outputs it is unknown if trajectories are referenced and ability of that software to work real-time.
Section 4 “Experiments”. I don’t see the reason why it couldn’t be “Result and discussion” section. The relevance of several mentioned datasets is questionable.
Section 4.3. Hardware specifications we have! It definitely fits “Materials and methods” section. And how about the software?
L398-401. These metrics need reference.
L465. “Figre” Typo?
Section 5. Is that really important to dedication the entire section to ablation study? That is forestry journal, not the machine learning journal.
L484. Which challenges exactly? You did not explain them. Not shown why existing solutions need re-engineering.
L487. “Mamba-structured”. You haven’t even explained or referenced what the “mamba-structure” is.
L512. It is really confusing. How it is supposed to reproduce results by the readers? How to check them independently?
Conclusion. I recommend to reject that manuscript and encourage authors to resubmit when it is done. Why wouldn’t perform data driven research sharing your source code with the others? Then proofread your own manuscript and show it to colleagues in the field for comprehension check. Such approach generally helps.
Comments on the Quality of English Language
L465. “Figre” Typo?
L189; 376; 393; 402; 410; 418; 432; 441; 451;463;468;476;. Orphaned heading? Where are their numbers?
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for Authors- The authors of the article probably forgot or inadvertently did not provide the literary sources that need to be entered when converting.
2. In Figure 1, I suggest improving and enlarging The t-th frame picture, The previous trajectories and results, because the essence cannot be seen.
3. In Figure 4, I also suggest improving the quality of the images mentioned.
4. I suggest adding a discussion where there is room to highlight the uniqueness of your research and the possibilities of its connection with practice.
5. In conclusion, I also suggest adding possibilities for continued research.
The article on the topic:
Time Series Correlation Optimization for Forest Fire Tracking for the special issue of Advanced Technologies for Forest Fire Detection and Monitoring consists of 17 pages of text, 6 main chapters, 9 figures, 7 tables, and the authors apparently forgot to assign literary sources to the scientific and professional texts used.
The agreement and overlap with the control texts is at the level of 11 percent, which I consider correct for the given issue.
The abstract, in terms of information, provides a robust basis for the research being presented. It is sufficient, however, it provides data on specific quantities, which I hope will be sufficient for the research, that the given percentages will have a sufficient basis for the research.
The introduction provides interesting theories and conclusions regarding the origin of fires and the detection of forest fires themselves, where, as I mentioned,
it is difficult to evaluate scientific facts, since the given conclusions are not substantiated, since the authors probably forgot to provide references to scientific research.
Research related to the detection of forest fires is processed in quite detail, but I will insist on reworking and supplementing the research regarding references. The introduction also provides sufficient information about the presented research, specifies the direction of the research and refers to research that has been carried out on a global scale.
In chapter 2 the authors describe multi object trackering and the modern current division into certain paradigms.
Various methods of observation and multitasking are described here, but I cannot evaluate the references, as I said, it is necessary to correct the cross-references to the research.
Chapter 3 - Methodology, is well-written, here I can state that the authors paid attention to the details of fire detection.
In Figure 1, I suggest improving and enlarging The t-th frame picture, The previous trajectories and results, because the essence is not visible.
In Chapter 3, they correctly noted and clarified the trajectories of the time-physical similarity model and the trajectories as such, which I appreciate.
For equation 2, I suggest including the derivation in the text and what the authors specifically meant.
I also suggest improving the quality of the images in Figure 4.
The Structure of the hierarchical association policy is correctly embedded in the context, which I also appreciate, as well as the Structure of the virtual trajectory generation module.
Apart from the above comments, I have no serious reservations about the chapter and I agree with the scientific substantiation of the given issue.
Regarding the experiment itself and chapter 4, I state:
The authors describe 4 datasets of images that characterize a certain depiction of forest fires.
The mentioned experimental settings are made at a high quality level and have a scientific output, which I also appreciate.
Table 1 shows a comparison of a sufficient number of models, which has a scientific basis and a coherent output.
The overall research seems quite complex, but of high quality, it needs to be studied sufficiently.
The individual chapters of the research are appropriately linked to each other and form a coherent document. The research is interesting and engaging for the reader.
What I would suggest is to reconsider the discussion, where the authors can propose a bridge with practice, as a possible use in solving forest fire detection.
The conclusion also reflects the overall article and describes and summarizes the overall research, where I would suggest adding the author's possible uses and connections to subsequent research and highlighting the uniqueness of the article.
Thank you very much, the article is of high quality, the individual chapters are processed at a scientific level and appropriately form a coherent document.
References need to be corrected.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThis article presents a novel approach to tracking fire in complex dynamic environments using temporal-physical similarity metric, Hierarchical Association Strategy, Virtual Trajectory Generation module. The authors argue that this approach increase tracking performance in previous research. Through a set of experiments, the authors provide evidences on the effectiveness of the proposed AO-OCSORT method
In my opinion, although the paper appears to be relevant to the field, it suffers from substantial issues and shortcomings that affects its potential research contribution. The following are the list of the identified shortcomings:
- The manuscript does not include a list of references and Discussion section. This is a big drawback of the article.
- Furthermore, both the Abstract and Introduction sections are poorly written and difficult to understand. Especcially Introduction sections should be rewritten for clarity, coherence, contributions of the work.
- The article's description of the customized FireMOT dataset is incomplete. The absence of crucial information about the dataset's structure, composition, collection method, and any preprocessing measures reduces the experiments repeatability and legitimacy. At least link or other source should be provided for this dataset.
- Formula 4 includes the parameter A, but there is no explanation of how this parameter is computed or obtained. All variables in formulas should be clearly defined and explained in the text.
- In Formula 3, it is stated that the parameters ∆, B, and C are derived from the Linear(Etie) layer, but Figure 3 shows that A, B, and ∆ are produced. This inconsistency leads to confusion and must be resolved to ensure clarity and correctness.
- The parameters C and D in Formula 5 are not defined, and there is no explanation of where they come from. Every symbol and parameter in the paper should be explicitly described.
- The description and interpretation of Figure 7 are too brief. A more thorough explanation is necessary to help readers understand what the figure represents.
- Table 2 shows that the ByteTrack method achieves the lowest FN value, yet the AO-OCSORT method is described as the best.
- The paper uses one set of evaluation metrics (HOTA, MOTA, IDF1, AssA, DetA) for most datasets, but a different set (MOTA, IDF1, FP, FN, IDSW) for the VisDrone dataset. No justification is provided for this inconsistency. Additionally, the metric IDSW is mentioned without any definition, which should be addressed in the evaluation section.
It is challenging to follow the writers' arguments and conclusions because of the manuscript's many grammatical problems and confusing sentence structures. The intended meaning is lost in a number of sections because of imprecise wording and inconsistent terminology.
I suggest having a professional editing agency or a native English speaker proofread the paper thoroughly. In order to fix the many small errors and raise the presentation's overall rigor and clarity, the authors should also thoroughly go over all of the formulas, notations, and text.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsCritical response on a revised paper entitled “Time Series Correlation Optimization for Forest Fire Tracking” submitted by Dongmei Yang et al to Forests
Firstly, I would like to thank authors for the thorough following to my advises. The revised version of manuscript describes computational framework in all necessary details, that let us to recommend published paper as a guide for forest researchers who learn implementing of UAVs for forest fires tracking. Also, quality of figures was improved and all necessary explanations were made.
However, there are minor issues that need to be solved.
Figure 1. Subpictures on figure 1 have to be numbered or marked with the letters and properly explained in the figure description, meaning aerial or land camera location, focal lenght, time of the day, fire spreading peculiarities. Same is relevant to the FireMOT dataset. Predominance of the certain image types may cause imbalance.
Figure 6. “ROI” is mentioned before introduced at line 284. Consider to expand it earlier?
Line 404. No data found. GitHub repository does not contain any data so far. When the code and data supposed to be placed there? It should be made be approval paper for publicaiton.
Conclusions. Minor issue need to be solved. Repetitive revision is not needed.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe authors of the manuscript have revised it based on the suggestions and comments provided, and sufficient efforts have been made to avoid misunderstandings. I consider the current state of the manuscript to be fully revised and recommend it for publication.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
