MS-FRCNN: A Multi-Scale Faster RCNN Model for Small Target Forest Fire Detection

Round 1

Reviewer 1 Report (Previous Reviewer 1)

Thanks for revision.

All my concerns have been addressed. I recommend for acceptance.

Reviewer 2 Report (Previous Reviewer 2)

I already reviewed this paper and mentioned the below comments, which the authors addressed. 

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

 

This paper proposed a two-stage object detection network for forest fire detection. Concerning the small target, authors designed a multi-scale feature extraction module. Moreover, a PAM module is incorporated to enhance the semantical correlations. Finally, the soft NMS algorithm is used to reduce the error deletion. This work is interesting, however, there are several issues that need to be addressed before recommending it for publication.

 

1. Essentially, the main scientific problem lies in the detection of small targets in forest fire identification. Therefore, more work on small target detection should be reviewed and discussed. The following references are suggested, DOI: 10.3390/rs14164065.

2. Move the non-original content of the proposed method to Preliminaries in section 2.

3. More comparative methods (SOTA) is encouraged in addition to baselines in experiments.

4. The proposed PAM is similar with CBAM. The necessary explanations are required for this point.

5. The English writing should be thoroughly revised.

 

 

Reviewer 2 Report

- In Figure 1, there is a big problem in data leaking. This is caused by the fact that the images in the dataset are taken from video frames, which result in multiple look-alike images and thus overfitting results. 

- There is no mention of the validation set on line 123, is it the same as the testing set?

- The writing style need to be greatly improved, for example, on line 119 "the data set is a dataset". 

-  Provide the precision-recall curves for the results. 

- The authors should have emploed cross-valiation for the evaluation method.

- What was the backbone model used?

- What was the value of the IoU used to produce the results.

- How many anchors and what was the location of detection heads?

- The choice of parameters and evaluation methods should be qualified with appropriate references, see similar studies utilizing Yolo (and faster RCNN mention in table 2 of the manuscript) can be cited so that to established the trustworthiness of the models and can provide reliability to baseline settings, see Detection of K-complexes in EEG signals using deep transfer learning and YOLOv3. Cluster Comput (2022). https://doi.org/10.1007/s10586-022-03802-0

- Separate the reference number from the text in the citation.

- The table of abbreviations is missing but required by the journal template.