FAUNet: Frequency Attention U-Net for Parcel Boundary Delineation in Satellite Images
Round 1
Reviewer 1 Report
The introduction and related works sections were disjointed and need to flow better. I didn't know why I was reading the article. What was the identified problem/knowledge gap. What are the disadvantages of previous studies in boundary detection. Why did you do this research and not use previous work? It was also a bit unclear if it was specifically agricultural boundary detection or general boundary detection (e.g. buildings, roads etc). I assumed field boundary detection.
The first paragraph was a bit fragmented and confused me to what point you were making. What’s your research problem? Is there a lack of knowledge in the literature on how to do instance segmentation for boundary detection? I didn’t think there was.
When discussing previous studies accuracy, it would be useful to quote. For example on line 41 “Xia et al and [7] and Garcia et al. [8], who successfully trained U-Net models on high-resolution images to accurately extract field boundaries and vineyard boundaries” – What was the accuracy? 75%, 90%? Perhaps something like this: “Xia et al and [7] and Garcia et al. [8], who successfully trained U-Net models on high-resolution images to extract field boundaries and vineyard boundaries to an accuracy of XX% and XX% respectively”.
You list many previous studies which have already done what you’re proposing and don’t mention why their method doesn’t work or their accuracies. Why is your study needed?
The methods were incomplete with some method contained within the results. The experimental design is a bit flawed with the validation conducted on the same sentinel path and I don't feel joining sentinel tiles are geographically distinct enough - perhaps some discussion justifying these distinctions regions and why the data is distinct enough (even with the same capture day) for this to be valid.
Referencing was inconsistent.
Limitations and future work details needs work.
See comments in attached PDF.
Comments for author File: Comments.pdf
Many grammar and syntax mistakes. This manuscript needs more proof reading to be publication standard. Some sections are good but others need work.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 2 Report
In this study, the authors proposed a straightforward yet highly effective method for boundary delineation that leverages frequency attention to enhance the precision of boundary detection. This approach, named Frequency Attention U-Net (FAUNet), builds upon the foundational and successful U-Net architecture by incorporating a frequency-based attention gate to enhance edge detection performance. Unlike many similar boundary delineation methods that employ three segmentation masks, this network employs two, resulting in a more streamlined post-processing workflow. The essence of frequency attention lies in integrating a frequency gate utilizing a high-pass filter. This high-pass filter output accentuates the critical high-frequency components within feature maps, thereby significantly improving edge detection performance. It showed good results as described in the paper.
The paper needs major revisions as follows:
- Clarify your main contribution in the Introduction section compared to existing methods. You know that many deep learning have been proposed to solve this problem. What is the newly presented in this paper?
-More comparisons are needed to more advanced deep learning methods, published in RS journal, as well as other journals.
- The method is hard to implement by the current description. It is so confused. Thus, sharing source codes is necessary to verify the reality of the application.
- The complexity of the compared approaches must be given.
- Improve the conclusion by addressing the limitations and future directions.
- English proofreading is needed, there are many typos.
English improvements are needed.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
Specify the Remote Sensing Dataset: Provide details about the remote sensing dataset used in the study, including its source, acquisition parameters, and any preprocessing steps. This will enhance the reproducibility of the results and help readers understand the data's characteristics.
Explain the Frequency Attention Mechanism: Elaborate on how the frequency-based attention gate works and its theoretical basis. Describe the significance of high-pass filtering in edge detection and provide visualization if possible to clarify the concept.
Compare with State-of-the-Art: Include a comprehensive comparison with existing boundary delineation methods, highlighting the strengths and weaknesses of FAUNet in relation to other approaches. Discuss the key differentiators that make FAUNet superior.
Provide Visual Examples: Include visual examples of boundary delineation results produced by FAUNet. This can help readers assess the quality of delineation and understand the practical implications of the method.
Discuss Computational Efficiency: Mention the computational resources and time required for running FAUNet. If it offers computational advantages over other methods, such as faster processing or lower resource requirements, emphasize these points.
Address Limitations and Future Work: Acknowledge any limitations of the proposed method, such as cases where it may not perform optimally. Additionally, suggest directions for future research or improvements to further enhance FAUNet's capabilities.
Please avoid citing sources that were published before to 2019. Cite current research that are really pertinent to your topic. The study also lacks sufficient citations. Another critical step is to compare the topic of the article to other relevant recent publications or works in order to widen the research's repercussions beyond the issue. Authors can use and depend on these essential works while addressing the topic of their paper and current issues.
Xu, K., Guo, Y., Liu, Y., Deng, X., Chen, Q.,... Ma, Z. (2021). 60-GHz Compact Dual-Mode On-Chip Bandpass Filter Using GaAs Technology. IEEE Electron Device Letters, 42(8), 1120-1123. doi: 10.1109/LED.2021.3091277
Chen, J., Wang, Q., Peng, W., Xu, H., Li, X.,... Xu, W. (2022). Disparity-Based Multiscale Fusion Network for Transportation Detection. IEEE Transactions on Intelligent Transportation Systems, 23(10), 18855-18863. doi: 10.1109/TITS.2022.3161977ZZFP33
Zhuo, Z., Du, L., Lu, X., Chen, J., & Cao, Z. (2022). Smoothed Lv Distribution Based Three-Dimensional Imaging for Spinning Space Debris. IEEE Transactions on Geoscience and Remote Sensing, 60, 1-13. doi: 10.1109/TGRS.2022.3174677
Li, Q., Song, D., Yuan, C., & Nie, W. (2022). An image recognition method for the deformation area of open-pit rock slopes under variable rainfall. Measurement, 188, 110544. doi: https://doi.org/10.1016/j.measurement.2021.110544
Fu, C., Yuan, H., Xu, H., Zhang, H., & Shen, L. (2023). TMSO-Net: Texture adaptive multi-scale observation for light field image depth estimation. Journal of Visual Communication and Image Representation, 90, 103731. doi: https://doi.org/10.1016/j.jvcir.2022.103731
Zhou, G., Li, H., Song, R., Wang, Q., Xu, J.,... Song, B. (2022). Orthorectification of Fisheye Image under Equidistant Projection Model. Remote Sensing, 14(17), 4175. doi: 10.3390/rs14174175
Zhou, G., & Liu, X. (2022). Orthorectification Model for Extra-Length Linear Array Imagery. IEEE Transactions on Geoscience and Remote Sensing, 60. doi: 10.1109/TGRS.2022.3223911
Zhuang, Y., Chen, S., Jiang, N., & Hu, H. (2022). An Effective WSSENet-Based Similarity Retrieval Method of Large Lung CT Image Databases. KSII Transactions on Internet & Information Systems, 16(7). doi: 10.3837/tiis.2022.07.013
Zhuang, Y., Jiang, N., Xu, Y., Xiangjie, K., & Kong, X. (2022). Progressive Distributed and Parallel Similarity Retrieval of Large CT Image Sequences in Mobile Telemedicine Networks. Wireless communications and mobile computing, 2022. doi: 10.1155/2022/6458350SZGC02
S. Aminizadeh et al., "The Applications of Machine Learning Techniques in Medical Data Processing based on Distributed Computing and the Internet of Things," Computer Methods and Programs in Biomedicine, p. 107745, 2023/08/09/ 2023, doi: https://doi.org/10.1016/j.cmpb.2023.107745.
Moderate editing of English language required
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 4 Report
I think it's a good article, it seems to me that there is a contribution to the delimitation of limits in satellite images. The detection of plots and the delimitation of boundaries is an important area today, especially in Smart Agro issues in applications of yield estimation and agricultural land management. The methodology is clear and well developed, the results are better compared to other techniques and the conclusions are supported by its results.
Author Response
Thank you very much for taking the time to review our paper. You comments are very generous and supportive. We thank you very much.
Round 2
Reviewer 1 Report
Thank you for the improvements. I believe this to be an exciting study however, I do not believe the study is yet at a standard to be published. There are still lots of editing errors is some sections. Both authors should thoroughly reread the manuscript so the quality is consistent.
The referencing is still inconsistent and doesn't always follow the style guide https://www.mdpi.com/authors/references.
The reference dataset https://collections.eurodatacube.com/denmark-lpis/readme.html is a vector dataset and doesn't contain the sentinel imagery as stated.
I do not believe training and testing on the same sentinel capture is an appropriate analysis even if conducted on a different part of the image. The authors should be able to download a sentinel image from another date to prove the model isn't overtrained on this image. Model accuracy tends to be high when run on the same image capture but some models accuracy falls dramatically when run on an image captured under different climatic conditions.
Please reread the manuscript as there are too many errors in the text. The English has improved compared to the first version.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
The authors did all comments. I have no more comments. This paper can be accepted.
Author Response
Thank you very much for your time. We appreciate all the notes and guidance you gave us. It elevated this work and made it much better.
Reviewer 3 Report
The paper was improved based on my comments.
Author Response
Thank you very much for your time. We appreciate all the notes and guidance you gave us. It elevated this work and made it much better.