Automated School Location Mapping at Scale from Satellite Imagery Based on Deep Learning

Round 1

Reviewer 1 Report

References appear twice at the end of the paper and conclusions are missing. 

I could not find detailed answers to the previous reviewer comments.

Author Response

Please see the attachment, thank you!

Author Response File: Author Response.docx

Reviewer 2 Report

Review report

 

Major issue:

The authors claim “Compared to the surrounding buildings, school structures are usually bigger in size, and the shapes vary from U, O, H, E, or L”. However, I still feel it is difficult to distinguish schools from other building types by using satellite imagery only, as their spectrum could be similar (as shown in Figure 1, it is hard to say there’s any uniqueness in these buildings identified as school). Nowadays, apart from satellite images, there are many geospatial big data available such as the OpenStreetMap that record school locations, and these data provide useful information for large-scale land use mapping including the educational use (Chen et al. 2021; Tu et al. 2021). Why not consider these additional data for mapping school locations?

 

Others:

Page 1, Lines 19-20. The authors claimed, “One of the key objectives of this work is to explore the possibility of having a global model that can be used to map schools across the globe.” However, I did not see any response in the abstract. According to the experimental results, regional models outperformed global models, indicating the statement was rejected. So what’s the point here?

 

Maybe add a figure of school classification maps in the manuscript.

 

References:

Chen, B., Tu, Y., Song, Y., Theobald, D.M., Zhang, T., Ren, Z., Li, X., Yang, J., Wang, J., Wang, X., Gong, P., Bai, Y., & Xu, B. (2021). Mapping essential urban land use categories with open big data: Results for five metropolitan areas in the United States of America. ISPRS Journal of Photogrammetry and Remote Sensing, 178, 203-218

Tu, Y., Chen, B., Lang, W., Chen, T., Li, M., Zhang, T., & Xu, B. (2021). Uncovering the Nature of Urban Land Use Composition Using Multi-Source Open Big Data with Ensemble Learning. Remote Sensing, 13

 

Author Response

Please see the attachment, thank you!

Author Response File: Author Response.docx

Reviewer 3 Report

Dear authors,

Your paper on “Automated School Location Mapping at Scale from Satellite Imagery based on Deep Learning”  focuses on a very relevant topic –the mapping of “missing schools”. I find the paper relevant and interesting to read. However, I would have several required improvements before publication:

• Introduction: this section is a mix of literature and a summary of your work. I would suggest removing the summary and focusing on an overview of recent publications and specify the research gap, and end this section with a short statement of the aim of your research. (I like Figure 1 - it provides good visual examples)
• Background: Provided in principle a good overview of deep learning and the reason for selecting a specific architecture – I would suggest adding the rationale for selecting CNN versus FCN.
• Material and Methods: I am missing (you submitted to Remote Sensing) an explanation about which images you used (did you use Google Earth images) or spectral images – which sensor and how different sensors might influence transferability? (please add to section 3.1). Also Figure 5-7 what is the data source?
• Results: Figure 11 – in particular the global model is not (well) explained!?
• Result: Figure 12: seem distorted – please improve!
• Discussion: I would suggest adding a short section how outputs can contribute to your main goal to bring schools on maps to support organizations such as UNICEF.
• I am missing a short Conclusion section

Author Response

Please see the attachment, thank you!

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

Overall, the authors have undertaken a comprehensive revision, and the current version is acceptable for publication in the Journal of Remote Sensing. 

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

1. Please add a diagram to show the proposed system architecture.
2. The authors should compare the proposed system with at least another published state of the art system from literature (or at least with a baseline method) in the term of precision, recall and F1. You may include a Table to depict numerical results (precision, recall and F1) of the comparison.
3. In addition, to improve your related work on building detection, you can cite the following papers.

 [1] Delassus, R., & Giot, R. (2018). Cnns fusion for building detection in aerial images for the building detection challenge. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops (pp. 242-246)..

[2] I. Grinias , C. Panagiotakis and G. Tziritas, MRF-based Segmentation and Unsupervised Classification for Building and Road Detection in Peri-urban Areas of High-resolution, ISPRS Journal of Photogrammetry and Remote Sensing, vol. 122, pp. 145-166, 2016.

[3] Ivanovsky, L., Khryashchev, V., Pavlov, V., & Ostrovskaya, A. (2019, April). Building detection on aerial images using U-NET neural networks. In 2019 24th Conference of Open Innovations Association (FRUCT) (pp. 116-122). IEEE.

Reviewer 2 Report

Dear authors,

I have not done the review, beacuse in checking the Turnitin similarity on internet it came out that you have to revise the paper regarding some non cited paragraphs. I'm adding the Turnitin report. The paper of which some of you are authors is even not in the reference list and in the text not cited. On web the similarity is 18%, in this attached report just 14%. The review can be done just after revision of the paper, with kidn regards

Comments for author File: Comments.pdf