A Machine Learning Approach for Identification of Low-Head Dams
Round 1
Reviewer 1 Report
The publication of the paper is very important and urgent task. The contents in the manuscript are well organized except of a few unknown comments.
Revision points;
1) Line 122; created >>> was created ; I am not a native English speaker. Please check.
2) Figure 5; What does the figure third from the left side mention?
3) Figure 10; What is the black inclined straight lines in the figure?
4) Figure 12; Please show clearly the Hydrofabric flowlines
Author Response
ALL of the comments from ALL of the reviewers are contained in this file.
Author Response File: Author Response.docx
Reviewer 2 Report
The Researcher has already cited Salvador thesis published under ref: https://scholarsarchive.byu.edu/cgi/viewcontent.cgi?article=10822&context=etd
Whereas there is a slight difference between his research and the already-published paper. The researcher has to address the novel approach of his work.
Explain the activation function in your work?
have you used the pooling layer? if yes than why?
What is the size of the feature map for a given input size image, Filter Size, Stride, and Padding amount
What is the size of your convoluted matrix? Explan it through numerically
Explain the terms “Valid Padding” and “Same Padding” in CNN.
List down the hyperparameters of a Pooling Layer.
Lastly it is strongly suggested to improve results and increase some latest research citation
Author Response
ALL of the comments from ALL of the reviewers are contained in this file.
Author Response File: Author Response.pdf
Reviewer 3 Report
This article addresses an interesting research topic for me. It reveals a machine learning approach to using high-resolution remote sensing data for the identification of low-head dams. Actually, it is a valuable attempt at identifying Low-head dams (LHD). However, the accuracy of the machine learning approach is not satisfactory in validation, as mentioned in the paper. Although the authors pointed out ways to improve their approach, I haven’t seen the results of the improved method. In my opinion, this article has not been completed yet. The paper is probably publishable but should be improved.
Additional Comments:
1. Express problems:
P.1, Line 11: “NLHD” should be written as “Non Low-head dams (NLHD)”.
P.12, Line 194-195, "The model classified images as LHD and was able to correctly identify 12 of 21 LHD locations," may be written wrongly. In Table 2, the number of LHD identified is 13.
2. Tables and Figures:
Table 1, format error, one more line was drawn.
Comments for author File: Comments.pdf
Author Response
ALL of the comments from ALL of the reviewers are contained in this file.
Author Response File: Author Response.docx
Reviewer 4 Report
See attached
Comments for author File: Comments.pdf
Author Response
Replies to BOTH reviewers 4 and 5 are found in the attached file.
Author Response File: Author Response.pdf
Reviewer 5 Report
In the manuscript the possibility of machine learning approach to accelerate the creation of the national inventory is testified. A machine learning approach is implemented to use a high-resolution remote sensing data with a Convolutional Neural Network architecture. It is concluded that the model achieved 76% accuracy in identifying low head dams and 95% accuracy identifying No low head dams on the validation set. Moreover, the developed model is used for the Hydrofabric flow lines. It is found that a high number of false positives and low accuracy due to the mismatch between Hydrofabric flowlines and actual water- ways.
The manuscript is very well written, research design is very good and results are quite useful.
The paper could be accepted in proper form. But, I have one little suggestion that could be implemented in the final draft i.e. It is better to declare the abbreviations first e.g. NLHD it is defined on line no. 126, and before that it is used at many instances.
Author Response
Replies to BOTH reviewers 4 and 5 are found in the attached file.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
Paper has been revised. The author has addressed many areas but yet few areas are not been addressed properly. It is suggested to add details methodology that may cover following areas:
activation function, pooling layer, size of the feature map for a given input size image, Filter Size, Stride, and Padding amount, size of your convoluted matrix? List down the hyperparameters of a Pooling Layer.
Author Response
Comment |
Response |
· activation function, pooling layer, size of the feature map for a given input size image, Filter Size, Stride, and Padding amount, size of your convoluted matrix? List down the hyperparameters of a Pooling Layer. |
· Activation functions used on the model are the following: o Relu, defined on section 2.3 of the paper manuscript o SoftMax, defined on section 2.3 of the paper manuscript · Polling layer: we implemented Max Pooling layers as specified on figure 8 and described on section 2.3 of the paper manuscript. · The size of the input image is defined as 128x128x3 as specified on section 2.2 of the paper manuscript. · We implemented a stride size of 1x1 and padding as valid as specified on section 2.3 of the paper manuscript. · The hyperparameters used for the Max pooling layer are pool size of 2x2 as specified on section 2.3 of the paper · As mentioned on section 2.3 of the paper manuscript, we implemented TensorFlow with Keras API to build the CNN model · We updated figure to show more details on the configuration of the CNN model
|