Comparative Research on Forest Fire Image Segmentation Algorithms Based on Fully Convolutional Neural Networks

Round 1

Reviewer 1 Report

Review report v1 – 1782619

 

This manuscript contributes to research into the segmentation of forest fire images. The topic is very current because many recent studies are based on the subject. However, this manuscript requires certain changes which are listed below.

 

Summary: It is necessary to refine the way of presenting more detailed information on the conducted research, to strengthen the importance of the impact of forest fires.

 

Introduction: This chapter is too long. It is recommended to shorten it by emphasizing the most important introductory contents. Writing quality is poor. The manuscript is difficult to read and understand.

 

Material and methods: This chapter requires a more thorough change to make it understandable. The methodology used in the manuscript is not clearly indicated. A more concise and clear presentation of this chapter is recommended.

 

Results: This chapter is too extensive. Part of the text can be used for discussion.

 

Discussion: The manuscript does not contain this chapter, which is inadmissible because the results obtained in this chapter should be compared with previous research. Refer to the numerous studies that have been done so far on the subject matter, which is mentioned in the summary. Be sure to prepare this chapter.

 

Conclusion: This chapter is too short. There are no clear conclusions on the purpose of the research conducted. The contribution of the research to the possible application in practice has not been shown.

 

 

The chapter: "scopes for future research" is not usually highlighted as independent. The authors should adhere to the uniformity of the Journal and the prescribed, usual structure, and it seems that there is a greater emphasis on this chapter than on the conclusions of the research.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper applies existing fully convolutional network architecture FCN, U-Net, PSPNet and Deeplabv3+ using two backbones (Resnet50, vgg16) for Forest Fire Images Segmentation. The main goal of this work is comparative research on remote sensing images of forest fires captured by unmannered aerial vehicles (UAVs). The subject of this paper is essential for wildfire studies. My major concerns of the paper are as follows:

(1)The reviewer suggests authors include some more recent work at least depending on the context of method. Here are some example references for your consideration.

 

i.        Ghali, R., Akhloufi, M.A. and Mseddi, W.S., 2022. Deep learning and transformer approaches for UAV-based wildfire detection and segmentation. Sensors, 22(5), p.1977.

ii.      Zhang, J., Zhu, H., Wang, P. and Ling, X., 2021. ATT squeeze U-Net: a lightweight network for forest fire detection and recognition. IEEE Access, 9, pp.10858-10870.

iii.    Shahid, M., Virtusio, J.J., Wu, Y.H., Chen, Y.Y., Tanveer, M., Muhammad, K. and Hua, K.L., 2021. Spatio-Temporal Self-Attention Network for Fire Detection and Segmentation in Video Surveillance. IEEE Access, 10, pp.1259-1275.

iv.     Shahid, M., Chien, I., Sarapugdi, W., Miao, L. and Hua, K.L., 2021. Deep spatial-temporal networks for flame detection. Multimedia Tools and Applications, 80(28), pp.35297-35318

v.       Harkat, H., Nascimento, J.M. and Bernardino, A., 2021, July. Fire Detection using Deeplabv3+ with Mobilenetv2. In 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS (pp. 4095-4098). IEEE.

vi.     Frizzi, S., Bouchouicha, M., Ginoux, J.M., Moreau, E. and Sayadi, M., 2021. Convolutional neural network for smoke and fire semantic segmentation. IET Image Processing, 15(3), pp.634-647

 (2) The innovation of this paper is low. Therefore, the reviewer suggests authors consider more advanced CNN backbone architecture and justify their work.

(3) In the experimental section, the reviewer suggests authors demonstrate the superiority of fire segmentation methods over forest fire image classification methods (Resnet50, vgg16).

(4) Since too small a test sample also influences the experimental results? So reviewer believes that adding more datasets for this analysis will be more effective.

(5) Authors must verify the real-time conditions and behavior of forest fire segmentation methods.

(6) The robustness of the forest fire segmentation methods under various conditions must be tested and included in the presented work.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

the authors improved their manuscript in all segments and made it acceptable for publication.