Sequentially Delineation of Rooftops with Holes from VHR Aerial Images Using a Convolutional Recurrent Neural Network
Round 1
Reviewer 1 Report
This paper presents a OEC-RNN network for detecting buildings with holes from remote sensing data. Some comments are:
1) Some papers published even in the same journal are missing such as:
Xu, Yongyang, et al. "Building extraction in very high resolution remote sensing imagery using deep learning and guided filters." Remote Sensing 10.1 (2018): 144.
Huang, Zuming, et al. "Building extraction from multi-source remote sensing images via deep deconvolution neural networks." 2016 IEEE International Geoscience and Remote Sensing Symposium (IGARSS). IEEE, 2016.
Weidner, Uwe, and Wolfgang Förstner. "Towards automatic building extraction from high-resolution digital elevation models." ISPRS journal of Photogrammetry and Remote Sensing 50.4 (1995): 38-49.
2) what about is a small size of a training set is utilised like in
Protopapadakis, Eftychios, et al. "Stacked autoencoders driven by semi-supervised learning for building extraction from near infrared remote sensing imagery." Remote Sensing 13.3 (2021): 371.
again published in the same journal.
3) some additional criteria can be exploited such as IoU
4) the computational cost should be given
5) the effect of the number of parameters should be given.
6) a statistical variation method should be analysed.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The main objective of the paper is a method for detection and localization of rooftops from buildings in very high resolution (VHR) aerial images. The paper includes a simple extension of the object-oriented delineation of rooftops with edges and corners using the recurrent neural network (OEC-RNN). Besides, several experiments shows interesting results in comparison with competitive methods. Thus, the contribution of the paper seems fair from a practical standpoint. In general, literal presentation of the paper, but there is room for improvement. Some details of the method needs more explanation. In addition, computational burden of the proposed method should be estimated. In summary, I consider the contents of the paper are potentially publishable, but the following specific issues should be addressed in a revised version of the paper.
- All the acronyms used in the paper should be defined the first time they appear in the text, except in the abstract, e.g., OEC-RNN, LSTM, VHR, ...
- Please include an explanation of the features used in the different steps of the method: definition, some examples, data distribution.
- An explanation of the tuning for each of the parameters of the method is required. For instance, (i) how erode and dilate operators, Figure 4, were implemented; (ii) page 5, line 182, kernel size 3, etc.
- It is know that neural networks have high computational load requirements. Please include a computational burden analysis of the proposed method in comparison with the other implemented methods in the paper.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
This work is similar to article published by the authors without high significant contribution(s).
Huang, W., Tang, H., & Xu, P. (2021). OEC-RNN: Object-Oriented Delineation of Rooftops With Edges and Corners Using the Recurrent Neural Network From the Aerial Images. IEEE Transactions on Geoscience and Remote Sensing.
The same method used in their previous work (OEC-RNN) is presented here with some modifications to consider building roofs with holes. This modification does not have the significant contribution to be published in the Remote Sensing journal. It could be suitable for a conference or a journal with a much lower impact factor. The OEC-RNN method is explained in details, three out of four comparison methods, one of the datasets and the evaluation parameters are described in their previous article and just referred in this paper.
Comments:
- Many abbreviations are mentioned without definition; please define each abbreviation since it is first time to mention, for example:
Line 16: “VHR” define in full
Line 17: what is this “Mask R-CNN”?
Line 67: what is this OEC?
Line 278: IoU?
And more …
- The word “both” was extensively used, please remove unnecessary ones, for example:
Line 10: Both semantic and instance segmentation are two types of commonly used methods for 10 high-resolution image building extraction >> Semantic and instance segmentation methods are commonly used for building extraction from high-resolution images.
Line 15 and Line 16: remove “both”
Line 21: “both one and multiple polygons” >> one or multiple polygons
And more …
- In the abstract, what is the study area its images specifications?
- Use more expressive keywords
- Line 12: resulting >> results
- Line 19: “and the external and internal” >> start a new sentence
- Line 20: What are these results that outperformed the state of the art methods?
- Line 20: shown >> showed
- Line 204: “Four metrics are used” what are they?
- Support the conclusions with results.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The authors have addressed all my previous concerns
Author Response
Thank you for your valuable comments!
Reviewer 3 Report
This work is similar to article published by the same authors:
Huang, W., Tang, H., & Xu, P. (2021). OEC-RNN: Object-Oriented Delineation of Rooftops With Edges and Corners Using the Recurrent Neural Network From the Aerial Images. IEEE Transactions on Geoscience and Remote Sensing.
The only new in this article is considering buildings holes. I don’t see it a significant contribution to be published in the Remote Sensing journal.
The same method used in their previous work (OEC-RNN) is presented here with some modifications to consider building roofs with holes. This modification does not have the significant contribution to be published in the Remote Sensing journal. It could be suitable for a conference or a journal with a much lower impact factor. The OEC-RNN method is explained in details, three out of four comparison methods, one of the datasets and the evaluation parameters are described in their previous article and just referred in this paper.
Author Response
Thank you for your valuable review comments!
