Fire and Smoke Detection Using Fine-Tuned YOLOv8 and YOLOv7 Deep Models

Round 1

Reviewer 1 Report

Comments for author File: Comments.pdf

Author Response

Thank you for taking the time to review our article, your comments are important for us to improve the quality of our paper. Below we present the responses to your comments.

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors have benchmarked a fire and smoke dataset on YOLOv7 and YOLOv8, where YOLOv8 outperforms the other model. However, there are major concerns regarding the manuscript:

1. The authors didn't propose a new dataset but rather utilized an existing dataset.

2. The authors didn't propose or modify a new model that is better in performance.

3. Keeping in mind the quality and level of the journal, this work is not suitable to be published in this journal.

Some Suggestions:

1. For this level of journals, authors either need to improve or propose a new model for the detection task; moreover, a comprehensive performance analysis is much needed to verify the efficiency of the new model or module in the existing model.

2. Another way is that the authors propose a new dataset for this task and can use existing models to benchmark and compare with other existing datasets.

3. The writing of paper needs extensive modifications. The authors have mentioned in the first paragraph of the introduction about transformers, which is irrelevent to the work. Moreover, the first paragraph should be broken down into multiple paragraphs.

4. State the contribution clearly to reflect the novelty of the work.

5. In cases where authors propose a new model, the details of the model should be presented in a comprehensive manner.

6. The experiment section can be further divided into subsections with an explanation of the dataset, hyper-parameters, results, and so on.

7. Comparative analysis with the previous works is much needed to verify the efficiency of the proposed work.

The language can be further improved to remove typos and grammatical mistakes.

Author Response

Thank you for taking the time to review our article, your comments are important for us to improve the quality of our paper. Below we present the responses to your comments.

Author Response File: Author Response.pdf

Reviewer 3 Report

Detecting smoke and fires in photos clearly has useful application and is a rapidly developing field of research and development. Unfortunately object detection is not an area that I have expertise in, so my review is not very substantial and probably reflects my ignorance of the subject matter.

A quick search reveals many recent papers on the topic. However, it seems that very few of them are referenced in this paper. Is this because this study occupies a specific niche, such that these papers are not relevant? In which case, a sentence or two providing the context would be helpful. Here are a few random examples:

Barmpoutis P, Papaioannou P, Dimitropoulos K, Grammalidis N. A Review on Early Forest Fire Detection Systems Using Optical Remote Sensing. Sensors (Basel). 2020 Nov 11;20(22):6442. doi: 10.3390/s20226442.

Azlan Saleh, Mohd Asyraf Zulkifley, Hazimah Haspi Harun, Francis Gaudreault, Ian Davison, Martin Spraggon. 2024.

Forest fire surveillance systems: A review of deep learning methods. Heliyon 10(1). https://doi.org/10.1016/j.heliyon.2023.e23127

Sathishkumar, V.E., Cho, J., Subramanian, M. et al. Forest fire and smoke detection using deep learning-based learning without forgetting. fire ecol 19, 9 (2023). https://doi.org/10.1186/s42408-022-00165-0

In the introduction, many specific models and datasets are referenced, including many acronyms with which the reader is assumed to be familiar. It is possible that fire management agency personnel may be interested in this material. A paragraph with more general information directed toward less-specialized readers would be helpful, again in setting the context and making the paper more approachable. For example, in fire management, the term "fire detection" generally means being the first to discover a forest fire; whereas here it used to mean identifying fire or smoke in a photograph. It would also be helpful to mention how some of the cited models are currently being used in fire management operations.

Comments on specific lines:

Line 6. "deep object detector" - is this correct/normal terminology?

16. "in risk" should be "at risk"

18. The fires in Australia in 2020 were indeed devastating, but this choice as an example may appear arbitrary. Note that wildfires killed 84 people in Greece in 2007, 88 in California in 2018, 101 in Hawaii in 2023, 104 in Greece in 2018, 114 in Portugal in 2017, 120 in Chile in 2024, and 173 in Australia in 2009.

21. In fact there are more small, low-intensity fires; even the large ones are not necessarily "sudden".

21. "for putting down" should be "to put out" or "to extinguish"

21. It is the fires that spread quickly that are difficult to control; the rest don't make it into the news.

22. remove "to act quickly and contain the development of forest fires if" - this goes without saying

24. detecting smoke is one way to detect forest fires, but not the only way

24. how about "it is essential for efficient fire management to know..."

25. "Sensor detection": suggest "remote sensing", which can include airborne sensors

26. "preventing" should be "detecting"

27. In most of Canada at least, only a small fraction of fires are detected by rangers on the ground. Most are detected by the public, pilots or passengers in passing aircraft, or airborne fire patrols. A few are detected via infrared satellite imagery.

31. This statement is incorrect. The primary sensors for detecting forest fires are not chemical or smoke detectors. The primary sensor used for fire detection is the human eye, either ground-based or airborne; followed by automated camera systems, typically tower-based, which may include IR imagery. There are numerous camera systems commercially available for automated or semi-automated fire detection (see for example FireTIR, AmpliCam, SmokeD, Nuvis, ADELIE, Stribor). I realize that the authors are probably familiar with these systems, but there is no mention of them in the paper.

64. "codec" should be "code"?

13-88. This is an extremely long paragraph. Be kind to your readers and break it up into smaller paragraphs.

73. "researches" should be "studies" or "research projects"; instead of "presented recently", consider "published recently"

75. "algorithm, their approach called..." should be "algorithm. Their approach is called..."

80. "damage coming from individuals..." should be "damage to individuals..."

94. "used to" should be "used with" or maybe "trained with"

94. Is it the "self-created" images that make ths approach "synthetic"?

141. This paragraph is about satellite-based images, and the only reference is to a paper about finding landing locations for UAVs. It seems that the entire field of fire detection via satellites is ignored. This field includes a large number of papers and widely-used operational systems monitoring fires in near real time (e.g. NASA FIRMS).

189. "The regression The loss..." appears to be a typo.

Figure 1, 5, 6. This is about forest fires, but there are many images here of urban and industrial fires. Would it make more sense to train a forest fire detection model using only images of forest fires? Also, all the images are visible; there is no mention of IR, even though the two are frequently effectively used together.

227. I'm curious about how many images contain fire or smoke, and how many do not.

260-272. It's not necessary to repeat so much of what is presented in table 1.

271. Figure ?? = Figures 2-4

277. "on night" should be "at night"

278. "with a high score of confidence reaches 0.83" should be "with a high confidence score of 0.83"

Figure 5. Looking at these pictures make me realize that specifying the platform first (e.g. traffic camera, tower camera, UAV, etc.) would allow more specific training of the model. In most cases the source of the image is known and may be a valuable piece of information informing the model.

282. should be "The proposed model exceeded the precision, recall, and mAP:50 in [40] by 6%, 3.8%, and 5.8% respectively."

316. "All the YOLOv8 models reaching a precision..." should be "The best YOLOv8 model reached a precision..."

326. A comment about application would be valuable. For what purpose, what platform, what kind of forest would this model be best suited?

Some suggestions on grammar and syntax are included in the general comments above.

Author Response

Thank you for taking the time to review our article, your comments are important for us to improve the quality of our paper. Below we present the responses to your comments. The other remarks for the corrections of the typos have been corrected in the manuscript.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

1. The authors claimed in their reply letter that they have benchmarked the dataset on YOLOv7 and YOLOv8, however, in the paper they say, "1. Presenting a fine-tuned YOLOv8 for smoke and fire detection in various locations. 2. Enhancing precision: The suggested method has the potential to enhance the precision of fire and smoke detection in forests, cities, and other locations when compared to traditional methods. A possibility to achieve this is by using the features of advanced deep learning algorithms like YOLOv8. These algorithms can be trained to recognize and detect specific characteristics of fire and smoke that can be challenging to identify using traditional image processing techniques," which seems contradictory.

2. For benchmarking, authors should evaluate the dataset on all the possible (and available) models in the literature that contribute to this domain.

3. This paper has no novelty whatsoever and is unfit to be published in this journal.

The language can be further improved to remove typos and grammatical mistakes.

Author Response

Thank you for taking the time to review our article, below is the response to your comment.

Author Response File: Author Response.pdf