VEPL-Net: A Deep Learning Ensemble for Automatic Segmentation of Vegetation Encroachment in Power Line Corridors Using UAV Imagery
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsI have carefully read the paper titled “VEPL-Net: A Deep Learning Ensemble for Automatic 2 Segmentation of Vegetation Encroachment in Power Lines 3 Corridors Using Drone Overhead Imagery” which discusses the use of several deep learning segmentation models for vegetation and power line mapping. The authors should address several issues as follows:
Abstract
There should be some brief statistical results to illustrate the superiority of the developed model.
Introduction
The main reference [2] is too outdated. What is the current situation?
It is written, “Multiple authors used semantic segmentation.” Such sentences are vague. You need to be more specific, and each sentence and paragraph must be connected.
The issue is not well described, and the significance of this study must be better described and discussed.
The writing needs to be significantly improved. There are excellent papers that discuss novel methods for improving segmentation tasks in the remote sensing field. Such studies must be included as the literature is too short.
Materials and methods
Both segmentation models of U-Net and DeepLab are outdated. There are much more advanced architectures for semantic segmentation. Why did you select these two models? You need to justify that, and it would be more helpful to compare your results with state-of-the-art segmentation models, such as TransU-Net++.
Results
In Tables, you need to use a dot (.) instead of comma (,). It looks too strange.
The results in the Figures are too close to each other, making it hard to interpret.
Discussion
The discussion section must be significantly improved. You need to compare your results with cutting-edge studies.
Conclusion
The conclusion section is too long and needs to be brief and concise. You need to conclude your findings and present future direction.
Comments on the Quality of English LanguageThe writing needs to be significantly improved.
Author Response
Thanks for the review
Author Response File: Author Response.docx
Reviewer 2 Report
Comments and Suggestions for AuthorsThis study utilizes deep learning for segmenting vegetation encroachment from UAV images. The topic is relevant, and the writing is generally easy to follow. However, I have a few major comments and suggestions for improvement:
1. Section 2.2 is more like a literature review than a methodology section, which can obscure the original contribution of this paper. The methodology section should focus on your unique contributions. Here are specific suggestions:
a) If you have modified the network architecture, it would be beneficial to include a diagram like Figure 3 for VEPL-Net and elaborate on the VEPL-Net architecture.
b) If you did not modify the architecture, I would suggest concentrating on explaining why UNet and DeepLab were chosen for this specific application. Also, delve into the transfer learning strategy (retain section 2.2.2). I would also remove the “VEPL-Net” name because it might imply a novel architecture, which isn't the case. I wanted to emphasize that even if you did not propose a new architecture, the contribution is still substantial if you explain network selection, transfer learning, loss function, and training strategy clearly.
2. Generalization and Applicability: Consider addressing the issue of how the proposed strategy might perform with different datasets. Can you offer guidance for someone aplying this strategy to their own dataset?
3. The IoU for power line in Tables 3 and 4 is considerably lower than that in Table 5. Is this correct? Please explain.
4. There are some typos throughout the paper (see examples below). Please check and fix.
a) Line 74: regions which are not connected “my” belong to the same object instance
b) Line 153: with the power line class being particularly dominant due to “his” geometry (lineal)
c) The decimal point should be represented with a period (".") rather than a comma.
Comments on the Quality of English LanguageMinor editing of English language required
Author Response
Thanks for the review,
Please find attached the responses.
Author Response File: Author Response.docx
Reviewer 3 Report
Comments and Suggestions for AuthorsThis manuscript focuses on automatic segmentation of vegetation encroachment in power lines corridors, which is of significance for energy societies. Moreover, the UAV images employed in this study can provide a huge prospect of applications. The proposed VEPL-Net shows outstanding performance in this study. The manuscript is technical sound and well orginized. Some issues should be suggested before further production.
1. More references should be expanded, especially the papers published in recent three years.
2. The number of Keywords is too many, 5 to 6 is best.
3. In Line 85 Page 2, authors introduced the atrous apatial pyramid pooling, the USPP (Automatic building extraction on high-resolution remote sensing imagery using deep convolutional encoder-decoder with spatial pyramid pooling), combined with Unet and ASPP, is suggested to be introduced.
4. More figures and ablation experiments are suggested to be added.
Comments on the Quality of English LanguageMinor editing of English language required
Author Response
Thanks for the review, Please find attached the responses.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThe results and discussion sections should be merged.
Comments on the Quality of English LanguageIt has been improved.
Author Response
Thanks for the review,
Please find attached the responses.
Author Response File: Author Response.docx