Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper is very interesting especially the authors treat a real problematic for Fire detection.

Some point Can be enhanced as follows:

The related work must include others researchers

The comparative study with the literature must be added

The Link and the appropriate code must be shown

 

Comments on the Quality of English Language

Some sentences must be enhanced.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Specific Comments

1. The novelty and originality of the authors’ proposal is questionable due to high similarity of approach with previous works especially the works of:

Chen G, Zhou H, Li Z, Gao Y, Bai D, Xu R, Lin H. Multi-Scale Forest Fire Recognition Model Based on Improved YOLOv5s. Forests. 2023; 14(2):315. https://doi.org/10.3390/f14020315

 

2. The author’s emphasis is on YOLOv5 for real-time, lightweight, and accurate detection.

Why did the authors not consider works done using YOLOv7 and YOLOv8?  YOLOv7 and YOLOv8 are more suitable for real-time object detection and more accurate than YOLOv5 and there are several works to validate that. For instance, YOLOv8 achieved a 98% detection accuracy in the works of:  

Talaat, F. M., & ZainEldin, H. (2023). An improved fire detection approach based on YOLO-v8 for smart cities. Neural Computing and Applications, 35(28), 20939-20954. There are so many other works.

Therefore, the choice of YOLOv5 instead of the more recent YOLOv7 or YOLOv8 for fire detection is backward. Authors should provide substantial basis for improving YOLOv5 to achieve their goal instead of using YOLOv7 or YOLOv8 and provide side-by-side performance evaluation of result.

 

3. Also, authors claimed lightweight of proposed model in terms of reduced complexity. Why did the authors not include model complexity evaluation metric in the result analysis section especially as it regards comparing AEGG-FD with the conventional YOLOv5? Authors should justify that the proposed model is lightweight and compare it with the YOLOv5 model series?

 

4. Several works have shown that YOLOv5 can achieve up to 85% mAp, on fire detection which is still not sufficient, and this led to the use of YOLOv7 and YOLOv8. Authors should check works such as

Miao, J., Zhao, G., Gao, Y., & Wen, Y. (2021, October). Fire detection algorithm based on improved yolov5. In 2021 International Conference on Control, Automation and Information Sciences (ICCAIS) (pp. 776-781). IEEE.

 

5. Authors only mention YOLOv5 without explicit mention of the category of YOLOv5. What series of YOLOv5 is the author comparing their work with? Is it YOLOv5s, YOLOv5m, YOLOv5x, etc. These are the different series of YOLOv5 which reflect the trade-off between accuracy, speed, and computational complexity of each category of YOLOv5 in detecting an object. Authors failed to show this in their work.

Authors can check the works of:

Yar, H., Khan, Z. A., Ullah, F. U. M., Ullah, W., & Baik, S. W. (2023). A modified YOLOv5 architecture for efficient fire detection in smart cities. Expert Systems with Applications, 231, 120465.

 

6. Authors novelty claim is faulted by their result analysis presentation. If the inference speed is increased because of lightweight which is indicated by reduced FLOPs and faster FPS (as shown in Table 6), what is the value of chosen YOLOv5’s GFLOP and FPS? The ablation study result in Table 6 only indicates the learning ability of the proposed model and it is not sufficient to validate proposed model’s superiority over related works.

 

7. Authors did not compare the performance of their work with existing state-of-the-art especially modified versions of YOLOv5 by previous authors to improve fire detection such as:

i. Xue, Q., Lin, H., & Wang, F. (2022). Fcdm: an improved forest fire classification and detection model based on yolov5. Forests, 13(12), 2129.

ii. Yar, H., Khan, Z. A., Ullah, F. U. M., Ullah, W., & Baik, S. W. (2023). A modified YOLOv5 architecture for efficient fire detection in smart cities. Expert Systems with Applications, 231, 120465. Authors should carefully consider these works and other recent works to substantiate their proposal performance and claim of novelty.

 

8. The first use of word before abbreviation and other typographical errors occurred on several occasions in the manuscript. For instance.

-Page 2, paragraph 3, line; “…regression-based method YOLO, which…” YOLO is not defined.

-Page 3, Related Works section, paragraph 2, “…with FPN [31] and PAN [32] …” FPN and PAN are not defined.

 

Authors should define terms before using the abbreviated version of it.

Comments for author File: Comments.pdf

Comments on the Quality of English Language

The English language is relatively okay

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The paper proposes a new convolutional neural network YOLO architecture for fire detection based on 4 basic changes: reconstruction of the backbone, improvement of the neck, optimization of the loss function and addition of the flame-center detection function. The result on a challenging dataset reveals a good tradeoff between accuracy and computational cost. I believe the paper will be ready for publication after a minor revision.

 

Specific comments:

 

- In equation (3), name of functions must not be in italic in order to not be misunderstood as the product of variables, and the asterisk commonly denotes convolution (not multiplication).

 

- In the last paragraph of section 4.3.1, the term “to the optimal solution” seems to be too strong, considering that there is no algorithm with optimality guarantee for the referred loss function.

 

- Define what the mAP subscripts 0.5 and 0.95 stand for, even though it is a common notation.

 

- Do not indent text after equations in the same sentence.

 

- The are some orphan titles and captions in the text which would better be avoided.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf