Deep Learning Approach for Paddy Field Detection Using Labeled Aerial Images: The Case of Detecting and Staging Paddy Fields in Central and Southern Taiwan
Round 1
Reviewer 1 Report
- In the abstract, training hyper-parameters should not be mentioned.
- The authors should have reviewed the manuscript before submission, see the line numbering intersecting with the paragraph text on page 3.
- Figure 2 is not a contribution of the paper and does not add new information.
- I blieve that important information should always be mentioned first. Thus, section 2.1 about the equipment is out of order.
- Regarding the different categories, Is it correct that the rice stage can be judged by the time of year?
- I light of Table 3, the image acquisition is not clear. How did you end up with thousands of images?
- What was the IOU value used in the results.
- Similar works utilizing RCNN should be discussed and cited, see Khasawneh, N., Fraiwan, M. & Fraiwan, L. Detection of K-complexes in EEG waveform images using faster R-CNN and deep transfer learning. BMC Med Inform Decis Mak 22, 297 (2022). https://doi.org/10.1186/s12911-022-02042-x
This would confirm the validity and popularity of the methods and the choice of parameters.
- The performance parameters are not defined.
- Did you perform testing or validation?
- Include the precision-recall curves for the results.
- The authors should have reviewed the manuscript before submission, see the line numbering intersecting with the paragraph text on page 3.
Author Response
We attached our replies.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This study proposed a method of dividing and recognizing the paddy area for aerial images. In the proposed study, the target data was divided into four growth stages. The proposed study is expected to alleviate the ongoing workforce-based work.
To improve the quality of the proposed study, it is recommended that the following be reinforced.
1.
The Confusion Matrix is a table used to evaluate the performance of classification models. For the accuracy analysis of the proposed study, it seems necessary to analyze the results using the confusion matrix.
2.
To prove the originality of the proposed research, a comparison process with the existing research is required.
This study proposed a method of dividing and recognizing the paddy area for aerial images. In the proposed study, the target data was divided into four growth stages. The proposed study is expected to alleviate the ongoing workforce-based work.
To improve the quality of the proposed study, it is recommended that the following be reinforced.
1.
The Confusion Matrix is a table used to evaluate the performance of classification models. For the accuracy analysis of the proposed study, it seems necessary to analyze the results using the confusion matrix.
2.
To prove the originality of the proposed research, a comparison process with the existing research is required.
Author Response
We attached our replies.
Author Response File: Author Response.pdf
Reviewer 3 Report
The paper is not particularly innovative, neither for the objective (paddy rice field mapping) neither for the methodology (commercial software used). So the authors should highlight in the first instance which is the new content of the study.
In the Section 2.3 the CMFI is used. Give a reference, considering that this index is not commonly used. Being a linear function of NDVI I do not see its added value, diversely from what it is stated in the conclusions.
In the Section 2.6 the authors should explain the meaning and the functionality of the mask RCNN procedure (backbone, ResNet and so on) which are not familiar concepts to every reader. The authors should inform the readers if the proposed procedure is available also in other software packages.
The section Results is not very clear. Explain what is the "loss" for Figures 8 and 9; Fig.7 is unclear and apparently it does not add information.
In the Discussion it is not clear which differences are occuring among the considered study areas and why. For example, it is not understood why the dice values in Tab.10 for Xingang are remarkably lower than for other areas. The testing presented in the figures 13-14-15 should give also some values and not just a visual comparison.
Author Response
We attached our replies.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The authors addressed my comments.
Reviewer 3 Report
The authors have improved the manuscript by adding clarifications and integrations to critical points raised in the previous version
