An Intelligent Wildfire Detection Approach through Cameras Based on Deep Learning
Round 1
Reviewer 1 Report
This manuscript presents an intelligent wildfire detection approach through cameras based on deep learning. Several comments listed below are suggested to improve this manuscript.
1．There are too many research backgrounds in the abstract. Abstract is supposed to reflect the most innovative and important part in the manuscript.
2．Linguistic competence needs to be highly improved, and the expression of academic ideas should rigorous and accurate. Check the manuscript thoroughly.
3．The method of wildfire detection using deep learning has been widely studied. What is the innovation of this manuscript?
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
In this paper, a video camera intelligent wildfire detection method was proposed, which was improved based on YOLOV5S architecture to achieve low model size, reduce computational complexity, maximize accuracy and detect data in real time.The structure of this paper and the content of their improvement are clear，There are a few comments to further improve the paper.
(1)Some sentences contain grammatical mistakes.
(2)Relevant research background needs to be improved in INTRODUCTION. The introduction for Convolutional neural networks is too long.
(3)Two of the fires listed in Table 1 are identical, Please check the manuscript carefully.
(4)The wildfires described in this paper are mainly forest and grassland. Is this method applicable to other areas?
Author Response
please see the attachment
Author Response File: Author Response.pdf
Reviewer 3 Report
The authors modify the YOLOv5 network and apply it to the task of fire and smoke detection. The article is structured well and is fairly easy to follow. However, the authors did not benchmark their modified model with a similarly published paper in Fire Technology:
Wu, Zongsheng, Ru Xue, and Hong Li. "Real-Time Video Fire Detection via Modified YOLOv5 Network Model." Fire Technology (2022): 1-27.
I recommended the authors revise the paper to refer to this work as it is the most similar to the presented study. The article's introduction and discussion should be updated to discuss the differences between the proposed approach and that presented by Wu et al. It would be beneficial to include Wu et al.'s model in the benchmarking in Tables 5-8 and the discussion.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 4 Report
The article is clear, well-prepared and interesting, but I have some suggestion to improve the paper.
1. Please define Kappa coefficient you mentioned on the 3rd page.
2. I think it is good idea to write few words about K-means algorithm or at least give some references.
3. In my opinion figures are too far from the place you first mention about them in the text. Some of them are even placed in the wrong chapter. I can not understand Fig. 2, it is not clear which is part a), b) and so on.
4. On the 11th page there is the following sentence: " The completed dataset is shown in Table 2". Is it correct? I have impression this is in another table.
5. Please check your manuscript again in respect of language. There are some spelling and grammatical mistakes.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
The authors sufficiently described the differences between their work and that presented by Wu et al. In the introduction. The authors chose to not benchmark their model with Wu's since the code was not readily available. While this may be sufficient to r publication, it would still be beneficial to benchmark against this model. If the timing for publication allows, I recommend the authors contact the corresponding author to request the code for benchmarking.
Author Response
请看附件
Author Response File: Author Response.pdf
