Mapping Buildings across Heterogeneous Landscapes: Machine Learning and Deep Learning Applied to Multi-Modal Remote Sensing Data
Round 1
Reviewer 1 Report (New Reviewer)
Comments and Suggestions for AuthorsThe paper presents a meaningful study on Mapping Buildings Across Heterogeneous Landscapes. Besides, A convolutional neural network was first trained to identify building candidates in LiDAR data. To better differentiate between true buildings and false positives, the CNN-based building probability map was then used, together with 400–2400 nm imaging spectroscopy, as input to a gradient boosting model. Simple vector operations were then employed to further refine the final maps. This stepwise approach resulted in detection of 84%, 100%, and 97% of manually-labeled buildings, at the 0.25, 0.5, and 0.75 percentiles of true building size, respectively, with very few false positives.The median absolute error in modeled building areas was 15%.
The logic and innovation part are clearly. However, there still some mistakes need to be improved.
1. what the difference between your sttudy and related study in recent years should be introduced more for stress the innovation of the paper out;
2. Line 105: In reference [12], some people have already adopted CNN to map the LiDAR data. Why you still adopted this method to do the same thing? Please give a detail reason;
3. Line 207, Figure 2: XGBoost also being used in the flowchart, will that method increase the computational complexity and slow down the maping speed?
4. Line 232: Why 799 trainable parameters were adopted in training process?
5. Line 310, Figure 4: The CNN maps seems more fluctuating than XGB maps, how to explain that in essence?
6. Line 349, Figure 5: the figure is too vague, it's better to redraw that;
7. LIne 391, Table 2: the recall value od proposed method is 0.85, it is not the best among three methods, you should give a brief description.
8. the format of the paper should be consisted with target journal, and some errors should be corrected such as in Line 175;
9. the language and grammer should be polished by professional teacher, some grammar mistakes should be corrected.
Comments on the Quality of English Languagethe language and grammer should be polished by professional teacher, some grammar mistakes should be corrected.
Author Response
Please see attached file
Author Response File: 
Author Response.pdf
Reviewer 2 Report (New Reviewer)
Comments and Suggestions for AuthorsThis paper presents the production of building maps using remote sensing data on Hawai'i Island. The contributions include training a neural network, combining imaging spectroscopy and gradient boosting to differentiate true buildings, refining maps through vector operations, and achieving high detection rates with minimal false positives. Experimental results demonstrate the effectiveness of urban planning, resource management, and disaster response.
The main problems in the introduction are:
1. When submitting for the first time, the revision mode should not be used.
2. "Heterogeneous Landscapes" is an essential keyword in the title. However, it is necessary to provide a clearer definition of what constitutes a Heterogeneous Landscape.
3. In the abstract, the authors mention "This novel integration of deep learning, machine learning". However, the usage of the deep learning method is not adequately explained in the Method section.
4. In the Introduction, you should clearly state the research problem or objective of the study. What is the gap in knowledge or the problem that you are addressing? What are the research questions or hypotheses?
5. The experiments should be improved. Numerous research studies have been conducted on the extraction and mapping of buildings using remote sensing. Therefore, it is necessary to supplement the section with a comparative analysis of recent research findings.
6. The discussion should incorporate an assessment of the model's transferability. Can it be adapted to other regions?
Author Response
Please see attached file
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report (New Reviewer)
Comments and Suggestions for Authors1. While it is common for CNN to consist of multiple convolutional and pooling layers, the presence of multiple hidden layers is not a necessary requirement for a neural network to be considered deep. The depth of a deep neural network is determined by the total number of layers it possesses, rather than solely the number of convolutional layers. In the case of CNNs, depth often refers to the number of layers between different convolutional layers, rather than the overall network depth. Thus, while a CNN can be classified as a deep neural network, this is not always the case. In the initial version, the Materials and Methods section primarily focused on describing CNNs instead of deep learning.
2. Regarding transferability, it is worth noting that remote sensing images often exhibit unique regional characteristics, which are commonly analyzed using deep learning methods. The CNN used in this paper is essentially a form of data fitting. It is necessary to consider the issue of transferability. For instance, Hawaii is not known for having densely built areas. While there are some cities and towns in Hawaii, the land use is primarily focused on preserving the natural environment and agriculture. It's quite different from supercity. As mentioned in the cover letter, there is potential for this approach to be extended to other areas with comparable data availability. However, it is crucial to define what qualifies as "similar data." For example, should we consider the remote sensing data from New York or Tokyo “similar” to that of Hawaii?
Author Response
Please see attached file.
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
Comments and Suggestions for AuthorsThe authors aim to produce building maps by jointly using CNN and XGBOOST model with the help of LiDAR and hyperspectral images. Overall, the quality of this paper is high, the experiments are sufficient, and the results can demonstrate the effectiveness of their method. In my view, the paper meets the publication criteria in Remote Sensing. However, before publication, I think the authors should address the following issues:
1. Please explain the motivation to choose XGBOOST model among the gradient boosting model.
2. Please check the citation in Line 188.
3. Please add the detailed information of reference 1, including journal title, volume number, doi, etc.
4. A deep literature review should be given, particularly some currently advanced deep learning models in multi-modal remote sensing. Therefore, the reviewer strongly suggests adding some related works in the revised manuscript, e.g., 10.1007/s11432-022-3588-0 10.1109/TGRS.2023.3267890.
Author Response
Please see attached file
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThis manuscript proposed a deep learning algorithm to map building in heterogonous area. Although the topic is interesting and ongoing but there are some comments that should be considered for further possible publishing:
- The novelty of the proposed method must be highlighted in the abstract.
- There are a lot of keywords, reduce them.
- Reorganization of introduction is recommended.
- Read the manuscript carefully. There are some sentences which are grammatically wrong!
- Problem statement and literature review is too weak. Authors have just explained the importance of building mapping in Hawai’i Island in addition to limited research in the field of building detection have been presented. They should specifically define the challenges of building mapping in their case study. Moreover, literature review must be updated and more classified researches added.
- Why hyperspectral and LiDAR data are used? How do they become complementary in building detection? Explain it.
- “feature engineering” is not common in remote sensing community. Authors can group the machine learning approach into traditional and deep-learning methods.
- Figures and Tables have too long caption! Correct them.
- How do you register hyperspectral imagery and DSM?
- What is the interpolation method to create DSM from point cloud?
- How do you handle difference in resolution of DSM and hyperspectral imagery?
- Add the flowchart of the proposed method in section 2.
- Authors mentioned that they use CNN to classify LiDAR data into building and non-building classes. What is the architecture of the CNN network? (they explained that it is slightly modified version of the UNET CNN architecture. What is the modification?). How many layers/parameters does it have? What are their layers (convolutional/pooling/batch normalization…)? Implementation setup can be added.
- Some thresholds are considered. How do you set them?
- Authors should add information about hardware configuration and the platform which they implement the proposed method in addition to computation complexity and time of training and inference.
- What will be the results if RGB images used instead of hyperspectral images.
- Methodology and results are mixed in Section 2. It is not required to explain the results in section 2.
- For better understanding, it is recommended to explains the problems of CNN results on DSM and the reasons of using hyperspectral images.
- It is not clear what is the output of XGBOOST model and how it fuses with results of deep learning. Because authors mentioned in section 1 that they use feature level fusion but it is not explained clearly in the methodology.
- Vectorizing step is very important. Explain more!
- According to the derived resolution, what is the accuracy of building boundary detection in ground system?
- There is not any accuracy table in the result section!!! (The accuracy of LiDAR data and hyperspectral separately and finally the results of the proposed method)
- Rewrite and reorganize the experimental results. Consider more tables and figures for this section.
- Moreover, consider following comments:
o The first sentence of introduction is too long, rewrite it.
o Table 1 is not necessary! It does not contain meaningful information.
o Use hyperspectral imagery instead of imaging spectroscopy.
o Line 115, “D” is incorrect.
o Figure 1 is simple and not necessary.
o Figure 2 has not enough quality.
o Check line 188!
o Figure 4 is not described and analyzed in the text.
Author Response
Please see attached file
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsMapping Buildings Across Heterogeneous Landscapes: Machine Learning and Deep Learning Applied to Multi-modal Remote Sensing Data
Dear Authors
The basic science of this paper is conducted in a good way and is of an appropriate standard. The author and his team write this paper according to journal scope and modern trends. I am glad to review this paper, because this manuscript is very relevant according to my research. In this study, the author described the production of maps of buildings on Hawai’i Island, based on complementary information contained in two different types of remote sensing data. The maps cover 3200 km2 over a highly varied set of landscape types and building densities. A convolutional neural network was first trained to identify building candidates in LiDAR data. Moreover, paper is not well-structured in methodology section. I have some major comments are given below.
Major comments
· The author should add couple of line of background of study in the beginning of the study.
· It’s better to modify the whole abstract. I am not satisfied
· At the end of the abstract provide what is significance of this research.
· Reduce keywords. Maximum 5 keywords
· The author used very old reference.
· Many statement needs references such as line 27-29.
· Line 127: The author add study area heading here.
· Modify figure 2 according to international journal criteria.
· Must add grids. Moreover the author (s) can use DEM, show elevation in different groups.
· Line 188: check reference here. There is a problem.
· Rewrite methodology section. That is very complex. The author should rewrite according to objectives.
· The author (s) also add flow chart of the whole study.
· Methods and methodology is mix here. Explain separately.
· Explain results according to objectives.
· Conclusion section is also missing.
I hope, the authors will modify research and resubmit in this journal.
Best Regards
Comments on the Quality of English Language
No comments
Author Response
Please see attached file
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsNo more questions.
Author Response
thanks
Reviewer 2 Report
Comments and Suggestions for AuthorsComments for author File: Comments.pdf
Author Response
please see attachment
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsI regret to inform you that your manuscript has been rejected due to your inappropriate response to the comments provided by me. I expressed dissatisfaction with your responses, which ultimately led to our decision. We appreciate your understanding.
Sincerely,
Author Response
please see attachment
Author Response File: Author Response.pdf
