Semantic Segmentation-Based Building Footprint Extraction Using Very High-Resolution Satellite Images and Multi-Source GIS Data

Round 1

Reviewer 1 Report

The manuscript titled “Semantic segmentation based building footprint extraction using high-resolution satellite images and multi-source GIS data” (Manuscript ID: remotesensing-432854) presents a procedure for building footprint extraction from WorldView-3 pan-sharpened imagery and GIS map data using semantic segmentation. The manuscript is well written, the novelty of the approach is explained clearly and the obtained results are challenging. My main concerns about the manuscript are:

1.      WorldView-3 is a very high resolution satellite image with 0.3m resolution. Therefore, please change the title as: “Semantic segmentation based building footprint extraction using very high-resolution satellite images and multi-source GIS data”

2.      As far as I understood from the text, the authors generated a random set (70% for training, 30% for validation) for each city during the analyses. It is better to produce at least three random sets for each case. Because, the obtained accuracies from the three random models will show that the accuracies are related to random sets or not. After applying these random sets, the model should be consistent.

3.      In figure 7 to 8, please explain the red and yellow polygons as in figure 5 and 6, instead of writing “… the same symbols will be applied to Figure 7 to Figure 9” in the figure 6 caption. You can put a small legend.

4.      In figure 10, it would be fine to show different color building boundaries for TP, FP, FN building areas for better understanding.

Author Response

Please check the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Originality of Subject and Recommendation

Although the chosen methodology is not new the topic and the overall approach seem to be interesting and the described workflow seems to be promising compared to other approaches. But the manuscript needs significant improvement before publication. The overall structure is clear but both research questions as well as concluding remarks related to these research questions are missing or at least not well described!

General Comments

A professional English language writing and grammar (which is currently an inconsistent mix of past, present and future) check is recommended. Several sentences need to be rephrased and it would be much easier to read through the paper afterwards. I just indicated a few examples below.

Detailed comments are summarised in the attached document!

Comments for author File: Comments.pdf

Author Response

Please check the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The problem of semantic segmentation for building footprint extraction is presented based on use of

satellite imagery and GIS datasets. Its a very interesting and mainstream problem with lots of activities going on.

Major comments:

1. I am surprised as to why the authors find it necessary to say this "

To the best of our

102 knowledge, no building extraction methods have been proposed and evaluated based on the

103 SpaceNet building dataset in existing journal publications.

"

That has no relevance to contributing new knowledge if there is any. Authors are recommended to take this sentence out. It appears in multiple areas throughout the paper. The sentence is also listed as on of the major contributions of this paper - which is bizarre. That contribution should be rephrased and rather highlight the incorporation of GIS data as per what challenge gets address by integrating satellite images with GIS data.

2. I am also surprised that the authors did not train a monolithic model that combines all data and compare with city level models. Could fine-tuning of the monolithic models provide additional performance boost to Khartoum datasets?

3. Section 3.3. first paragraph requires rephrasing for clarity. Reference to probability maps is made in several sentences and yet pointing to difference inputs. It's easy to confuse the reader. Not sure if those first two paragraphs from section 3.3. are meant to fo with Figure 3. Figure 3 requires some additional information to tie it to the narrative then e.g. show image sizes for each stage.

Section 4.3. Authors claim model generalization but do not provide enough details as to the context or characteristics of the data being tested e.g. new city samples etc.

Section 5.1. Please take out:

To the best of our knowledge, no building extraction methods have been proposed and evaluated based on the

SpaceNet building dataset in existing journal publications.

Contributions:

Integrating GIS data with satellite images is a novel idea.

Experiments are conducted with lots of variations

Minor grammar erros:

paper require proof reading by an english first language speaking professional.

Metric --> metric

Author Response

Please check the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The manuscript has significantly improved by the implementation of reviewers recommendations. I still strongly recommend English proof reading before publication. 

Author Response

Many thanks for your valuable comments. We submitted our paper to MDPI English Editing Service. The editors have made extensive English editing to our paper. We also proof read the revised paper and each correction carefully. We believe this manuscript has been significantly improved to a level suitable for reporting research in this journal now. Please find each revision to this article highlighted in our submitted manuscript.

Reviewer 3 Report

Main text has been improved in revised paper and authors responded to my concerns.

Author Response

Many thanks for your comments. As the editor and the second reviewer suggested that our manuscript should undergo extensive English editing, we submitted our paper to MDPI English Editing Service. The editors have made extensive English editing to our paper. We also proof read the revised paper and each correction carefully. We believe this manuscript has been significantly improved to a level suitable for reporting research in this journal now. Please find each revision to this article highlighted in our submitted manuscript.