Automatic Extraction of Martian Subsurface Layer from Radargrams Based on PDE Denoising and KL Filter
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsReview of “Automatic Extraction of Martian Subsurface Layer of radar images Based on PDE denoising and KL filter” By Xin Shu and Hongxia Ye.
Summary:
The SHAllow RADar (SHARAD) onboard Mars Reconnaissance Orbiter (MRO) is currently the most highest resolution radar sounding instrument for acquiring internal information of Martian poles. The observations reveals tens of horizontal reflectors, which are believed to records the global climate change during late Amazonian period. Because there are a huge amount of SHARAD observations and developing of an automatic reflector extraction algorithm is urgent. In this paper, the authors developed an automatic algorithm for extraction of subsurface reflectors in SHARAD radargrams. In this method, a fourth-order partial differential equation is used for radargram denoising and signal-to-noise ratio enhancement, and then Kullback-Leibler is employed to extract subsurface reflector. The authors compared their results with the method in previous studies, showing that the proposed method can produce more complete and accurate extraction results. The proposed method in this paper can improve reflector detection efficiency and can be used in subsequent research of Martian polar layer deposits.
In generally, the paper is technical correct, and the conclusions are well supported by the results. The paper is clear and well written. I have only a few minor comments/suggestions. If the authors can address these comments, I would suggest its publication in Remote Sensing.
Here are my specific comments:
P4. Line160-163. Please justify why normalization is needed here. In Figure 3, background noise is significantly reduced after brightness adjustment. Is this due to normalization or (1)? Please clarifiy this.
P5. Line237-240 and Figure 5: Which picture does figure (a)(b)...(f) correspond to? Please label all the subfigures which helps reader to understand the text. Why does the region of the data change after PDE diffusion denoising?
The definition of some parameters are not clear, please check section 3 and explain these parameters. Here are two examples: P11. Line338-340: Are Wa and Wb empirical parameters? How to choose the value of these two parameters? P12. Line352: Again, why is the threshold 50? Please clarify how to set the threshold value.
P14. Line401-407, SPLD results in the upper level continuity are not very good. Is it the problem of the orbital data, or is it possible that its own poor continuity affects the identification and classification of KL? Please clarify this.
L57–58 I suggest the authors give a short review on previous studies of reflector extraction and then justify why there is a need for developing a new method.
Figures. For all the radargams, I suggest the authors add a scale bar (both vertical and horizontal).
L2. Radar image: should be radargram; PDE and KL, please give the full name.
L9. Reserves should be reservoir
L10. layer detection should be reflector detection?
L16. KL give the full name; L61 AOS ; L73, PD4.
Comments on the Quality of English LanguageThe writing is clear and good, but there is still room for improvement.
Author Response
Dear Reviewer:
Thank you for the helpful comments. We carefully revised the manuscript according to the review opinions. All changes in the updated manuscript are indicated with red color. Below is the one by one response. Please kindly consider it for review.
Best regards!
Xin Shu, Hongxia Ye
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsReview of:
Manuscript ID: remotesensing-2831935
Type of manuscript: Article
Title: Automatic Extraction of Martian Subsurface Layer of radar images Based
on PDE denoising and KL filter
Authors: Xin Shu, Hongxia Ye *
Satellite Missions for Earth and Planetary Exploration
-------------------------------------------------------------
I recommend "Accept after minor revision (corrections to minor methodological errors and text editing)" subject to the authors addressing the following comments to the satisfaction of the Academic Editors.
-------------------------------------------------------------
Does the introduction provide sufficient background and include all relevant references? Yes
-------------------------------------------------------------
Are all the cited references relevant to the research? Yes
-------------------------------------------------------------
Is the research design appropriate? Yes
-------------------------------------------------------------
Are the methods adequately described? No
In line 164 equation (2) is said to remove "non-zero values". I don't see how equation (2) does this, and do not understand why one would want to remove non-zero values.
The labels (a)-(d) are missing from Figure 3. In line 170 it is implied (by the order) that the upper panels of Figure 3 correspond to orbit 00520501, but comparison with later figures indicates that the upper panels are for orbit 00505101. The uncertainty of which orbit is which is compounded by the absence of labels (a)-(f) in Figure 5.
In reference to equation (9), "i" and "j" are defined in line 195 as row and column numbers, but "N" is defined in line 219 as the number of grayscale levels. Thus the sum over row and column numbers up to "N" does not make sense. Should different symbols be used for the number of rows and columns in equation (9)?
In Figure 4(b) a local variance is plotted for iterative denoising of the area outlined by the blue box in Figure 4(a). At first sight I could not see any structure whatsoever within this box, so could not understand how it could have any variance. However, when I magnified it considerably, I could then see vary faint blobs. It would be helpful if the authors explained this in the text.
The upper left and centre panels of Figure 5 seem to be equivalent (in description) to the upper panels of Figure 3, but have much lower resolution. I do not understand why there are these differences in resolution, so do not understand which original data are used for the denoising process.
In the absence of any definition of the vertical scale in Figure 6, I assume that the largest peak is due to the reflection from the surface. If this is the case, then it seems there is a depression in the surface for profile B. However, in Figure 5 (upper left) the surface seems to get slightly higher in a progression from profiles A to D. Hence I do not understand the position of the surface reflection in the second profile in Figure 6. Also in Figure 6, there appears to be a feature that is descending in the sequence of profiles A to D, while I cannot see any corresponding feature in Figure 5 (upper left panel). The vertical scale in Figure 6 should be properly defined and the above points explained, so that the reader can understand the method here.
The red dots in Figures 6 and 7 are not defined anywhere in the text or figure captions. Thus the method being applied is not clear.
The details of the "visual recognition" method (line 381) should be given.
There are deficiencies in English expression that make it difficult to work out the method:
- line 162: should "expand the contrast of the light and dark areas" be "expand the contrast between the light and dark areas"?
- line 229: The use of the word "also" suggests that something is missing before this sentence.
There are technical terms that are not defined or referenced:
- line 73: PD4
- line 130: state transition probabilities
- line 254: simulated clutter simulation image
- line 288: sequence frame rates
- line 300: peakdet function
- line 333: sub-ice features
- line 434: Arctic and Antarctic
Some acronyms are not defined on first occurrence:
- line 16: KL
- line 61: AOS
-------------------------------------------------------------
Are the results clearly presented? No.
In Figure 5, the PDE diffusion denoising is plotted on an expanded scale relative to the original image and the Linear Brightness Adjustment. This makes it difficult to assess the results of the denoising. The 3 upper panels in Figure 5 should be plotted on the same scale so that the reader can assess the results of the denoising process.
In the original (top left) I can see only 3-5 layers, so it is hard to believe that the brightness adjustment could produce the approximately 12 layers in the top centre panel, or that the PDE diffusion denoising could then produce the approximately 20 layers in the top upper right panel. However, if the top right panel is assumed to have been calculated from the data in the upper panels of Figure 3, then the results are believable.
The results in Figure 9 should be discussed, e.g. for orbit 00505101 there are 24 layers in green markings, compared to only 4-5 layers visible in figure 5 (upper left). Again, were the data in Figure 3 used as the original, rather than those shown in Figure 5?
When I first looked at Figure 11 I could not identify the "red part". When I magnified it considerably I saw the pinkish-brown blobs that I assume are the "red part". It would be helpful if the "red parts" could be plotted in red.
There is a contradiction between the text and caption for Figure 11, in that the Elysium-Utopia Region is referred to in line 457, but is not referred to in the caption.
In line 459 there is reference to "locations indicated by arrows", but I cannot see any arrows in Figure 15. I therefore cannot assess the results.
-------------------------------------------------------------
Are the conclusions supported by the results? Generally yes.
However, there is nothing about training neural network models in the body of the paper, so this shoud not be included in the conclusions.
Comments for author File: Comments.pdf
Moderate editing of English language required:
There are many cases of inappropriate words, bad grammar, and missing spaces. I have highlighted these in green in the annotated version of the manuscript attached.
Author Response
Dear Reviewer:
Thank you for the helpful comments. We carefully revised the manuscript according to the review opinions. All changes in the updated manuscript are indicated with red color. Below is the one by one response. Please kindly consider it for review.
Best regards!
Xin Shu, Hongxia Ye
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsDear authors,
Thank you for submitting this paper. It certainly includes worthy results to be included in Remote Sensing Journal. Very little edits are required. Please, find the document attached.
Comments for author File: Comments.pdf
Author Response
Dear Reviewer:
Thank you for the helpful comments. We carefully revised the manuscript according to the review opinions. All changes in the updated manuscript are indicated with red color. Below is the one by one response. Please kindly consider it for review.
Best regards!
Xin Shu, Hongxia Ye
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments and Suggestions for AuthorsGeneral notes:
· All of the figures with radar images need scale bars.
· There should be a map and/or general discussion of the relevant geology in the polar regions. Without these, it is not clear to the reader where these radar images were taken.
Line Edits:
Lines 19-21: What does “good performance” mean relative to the other methods? It would be helpful to quantify the difference between your method and others to show the level of improvement in terms of missed detection rate and connectivity between layers.
Lines 26-27: Why do we care more about the sub-surface at the poles than in other regions on Mars? It’s not clear from the paper what makes the polar regions more interesting from a scientific standpoint.
Line 61: Define the acronym for AOS.
Line 232: It is unclear why the number of iterations was set to 7. It makes sense that more than 7 iterations will result in decreased sub-surface layer detection, but what makes the 7th iteration more optimal than the 6th or 5th? There needs to be quantitative justification for why the 7th iteration is necessary/better.
Figure 4: On the y-axes of B and C, what does metric value refer to?
Figure 5: The individual panels should be labeled a, b, c, etc.
Figure 6: This figure is not publication quality. There are no axis labels and the caption does not describe what is being shown in the figure. There is no indication of what the red dots represent or what the blue lines are measuring.
Figure 7: See comments for Fig. 6. Also, it is impossible to tell what effect the threshold filtering has had on this dataset when compared to the nonfiltered data in Fig. 6. The two figures look identical when placed side by side.
Comments on the Quality of English LanguageSee comments above.
Author Response
Dear Reviewer:
Thank you for the helpful comments. We carefully revised the manuscript according to the review opinions. All changes in the updated manuscript are indicated with red color. Below is the one by one response. Please kindly consider it for review.
Best regards!
Xin Shu, Hongxia Ye
Author Response File: Author Response.pdf
