Offshore Bridge Detection in Polarimetric SAR Images Based on Water Network Construction Using Markov Tree
Round 1
Reviewer 1 Report
The paper "Offshore Birdge Detection in Polarimetric SAR Images Based on Water Network Construction Using Markov Tree" attempts to describe a Markov Tree method to detect waterways in Pauli decomposition images from polarimetric SAR, then detect bridges.
This work suffers from several defects, keeping it from being a good, useful, and publishable paper.
1. First, there are several grammatical issues and several misspellings in the paper that should be addressed.
2. The authors assert that it is difficult to detect bridges in SAR due to the inherent speckle noise, but give no justification for this statement. Bridges, as seen against water, should have a very high SNR (including several different bounces off of the structure and onto the water and back to the sensor). The water (provided it is not highly choppy) should have a significantly lower SNR. So why need a fancy Markov Tree technique to find the water and bridges? As is the case for many feature extraction techniques, there is probably some really good reason for this technique, but there was little to no explanation in the Introduction. This needs to be better addressed.
3. The authors give absolutely no justification as to why they prefer polarimetric SAR over single-pol SAR. Since fully polarimetric SAR is hard to come by (many on-orbit sensors can only do single-pol SAR), the authors need to justify why this technique is suitable only for polarimetric SAR. There is no comparison with using single-frame VV or HH imagery. What is it about the polarimetry that makes this successful? Is it only successful with polarimetry or is there value in using only one pol configuration (VV or HH or VH/HV)?
4. The authors show no examples of bridges in SAR imagery, which might help the reader understand why this Markov Tree technique is needed.
5. The authors assert that "[t]he problem of the method mixing DEM and SAR data is that it is difficult to synchronize the results from DEM and SAR data". Really? Why? This statement is technically prejudicial without some discussion/evidence to back it up. A lot of DEMs are *created* from SAR data, so why is it so difficult to use these products together? Justification is definitely needed here.
6. There are interesting papers listed in the references. Ref [4]-[9] talk about various bridge extraction methods. Are any of them deep learning based? Has deep learning even been tried? This seems like a natural fit for CNNs: to find bridges in the midst of other objects.
7. The entire technical section (Section 3) that describes the trees, the probabilities, and the energies is very, very difficult to follow. For example, the authors define L as "the look", but never say to what this corresponds. Another example, Eq 5 is presented with no justification as to why it is defined as it is. Same for Eq 20. Figure 3 is too small, making it difficult to see what's going on as it is being described in the paragraph above the figure. The entire scheme could use a diagram to show each of the attributes (like l_L and l_W and P). In Eq 14, there is no discussion on how the priors, p(K), are found (is there a dataset used to define this?). Another example: the discussion on the Douglas-Peucker (DP) splitting and merging algorithm is very difficult to follow; perhaps a flowchart or diagram to help the reader understand. In the end, I don't think there is any easy way that someone could replicate the authors results based on the descriptions given in Section 3.
8. In Section 4, there is no comparison with other bridge extraction algorithms, nor is there a discussion as to why this method is better than other methods.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The topic is interesting and well writing technically after reading it some minor comments and questions are as follows;
- What is the main input of this research please clearly mention it.
- The output layout does not have scale, north arrow and legend please add.
- Please shortely write about preprocessing of PolSAR image.
Conclussion not well supported the results.
Author Response
Dear Reviewer 2:
Thank you for your valuable comments on our manuscript. After carefully studying your comments, we have revised the manuscript. Some clarifications for your comments are listed as follows.
[Reviewer 2 Comment 1]:
What is the main input of this research please clearly mention it.
[Response to Comment 1]:
Thank you for your valuable suggestions. The main input of the proposed method is the coherent matrix T, which is obtained from single look quad-polarization SAR data acquired by RADARSAT-2 sensor over Singapore region in 2013 and Lingshui, Hainan province, China region in 2014 and Multi-look quad-polarization SAR data acquired by TerraSAR-X sensor over Singapore region in 2014. First, the input coherent matrix is used to carry out the water-land segmentation using level set segmentation method. The sentence “When the proposed method is carried out, the coherent matrix (directly transformed by the Sinclair matrix for single look data) is calculated as the input data.” is added in the data description of Section 4.
[Reviewer 2 Comment 2]:
The output layout does not have scale, north arrow and legend please add.
[Response to Comment 2]:
Thank you for your professional suggestions. The scale, north arrow and legend were added in Figure 7 of the new manuscript.
[Reviewer 2 Comment 3]:
Please shortely write about preprocessing of PolSAR image.
[Response to Comment 3]:
Thank you for your professional suggestions. There isn’t any preprocessing of PolSAR image except the coherent matrix transformation from the Sinclair matrix. We directly input the coherent matrix into the proposed algorithm. The input data is first segmented by level set segmentation method.
[Reviewer 2 Comment 4]:
Conclussion not well supported the results.
[Response to Comment 4]:
Thank you for your valuable suggestions. Conclusions were revised to more matched the results in the new manuscript.
Yours sincerely,
Chun Liu
July 30, 2022
Reviewer 3 Report
The article was well written and makes a significant contribution to the scientific community.
I have only two questions for the authors to answer:
1 – Line 94: Could you explain better the reported problem about synchronization between DEM and SAR data?
2 – Line 470: What is the temporal correspondence between the set of SAR image data (RADARSAT-2 and TerraSAR-X) used and the reference data (Google Earth Map) that was used for validation?
Author Response
Dear Reviewer 3:
Thank you for your valuable comments on our manuscript. After carefully studying your comments, we have revised the manuscript. Some clarifications for your comments are listed as follows.
[Reviewer 3 Comment 1]:
Line 94: Could you explain better the reported problem about synchronization between DEM and SAR data?
[Response to Comment 1]:
Thanks very much for your professional comments. The problem about synchronization between DEM and SAR data is that the resolution of DEM is usually 10m, 20m or 30m but the resolution of SAR is usually lower than 5m. If the two kinds of data are not a same source, it is difficult to synchronize. The problem was revised as “the inability to consider manmade features using DEM [22] will also cause the inaccuracies of the regions of the manmade features using the mixture data” according to the reference [22].
[Reviewer 3 Comment 2]:
Line 470: What is the temporal correspondence between the set of SAR image data (RADARSAT-2 and TerraSAR-X) used and the reference data (Google Earth Map) that was used for validation?
[Response to Comment 2]:
Thank you for your instructive suggestions. The scale, north arrow and legend were added in Figure 7 of the new manuscript.
Yours sincerely,
Chun Liu
July 30, 2022
Round 2
Reviewer 1 Report
In the second review of "Offshore Bridge Detetion in Polarimetric SAR Images Based on Water Network Construction Using Markov Tree", the authors appear to have addressed most of the reviewer's concerns from the first review. However, there are just two minor issues that need addressing:
1. In the Introduction, Lines 31-34, the authors added a figure to help illustrate the difficulty of finding bridges. The issue os "scattering interference" is well noted and the authors present an excellent example in Figure 1. However, the argument that it is "not a simple procedure due to the inherent speckle noise of SAR" is not well supported. "Region 1" (which needs to be better indicated, as the blue against the blueish-purple image is hard to see) is a good example of how imaging geometry and bridge construction (metal versus wood) may play a factor in bridge detection. Perhaps adding the following to the sentence: "However, it is not a simple procedure as the imaging geometry may play a factor in significant reduction of the radar returns from the bridge so that it is difficult to discriminate between bridge and speckle noise due to the water (region 1 in Figure 1). Additionally, the scattering interference caused by strong scatterers along the coast, and the diversity of bridge and coastal terrain morphologies [1] (region 2 in Figure 2)."
2. From Comment 5 in the previous review, the authors added Reference 22, which is a good reference. However, the sentence in Lines 96-98 makes no sense. The "inability" does not cause "inaccuracies". Instead, the authors might consider: "The problem of mixing DEM and SAR data is that the DEM data is often inaccurate or incomplete [22], particularly when observing waterways," making it difficult to fuse the two data types for an effective water detection technique."
3. Finally, when the reviewer mentioned that "[t]he entire scheme could use a diagram to show each of the attributes", the idea was to produce a flowchart that showed each step in the process, along with each attribute/equation that is used or affected by that step in the process. Instead, the authors just generated a table of terms in Appendix 1. While this is useful, it is not what the reviewer was anticipating. Additionally, there are so many terms to keep track of in this technique, the reviewer wonders how appealing such a complicated technique will be to other researchers.
Author Response
Dear Reviewer 1:
Thank you for your valuable comments on our manuscript. After carefully studying your comments, we have revised the manuscript. Some clarifications for your comments are listed as follows.
[Reviewer 1 Comment 1]:
In the Introduction, Lines 31-34, the authors added a figure to help illustrate the difficulty of finding bridges. The issue os "scattering interference" is well noted and the authors present an excellent example in Figure 1. However, the argument that it is "not a simple procedure due to the inherent speckle noise of SAR" is not well supported. "Region 1" (which needs to be better indicated, as the blue against the blueish-purple image is hard to see) is a good example of how imaging geometry and bridge construction (metal versus wood) may play a factor in bridge detection. Perhaps adding the following to the sentence: "However, it is not a simple procedure as the imaging geometry may play a factor in significant reduction of the radar returns from the bridge so that it is difficult to discriminate between bridge and speckle noise due to the water (region 1 in Figure 1). Additionally, the scattering interference caused by strong scatterers along the coast, and the diversity of bridge and coastal terrain morphologies [1] (region 2 in Figure 2)."
[Response to Comment 1]:
Thanks very much for your professional comments. We added the sentence you given in the first paragraph of Section 1. The sentence is so wonderful to describe the problem of bridge detection.
[Reviewer 1 Comment 2]:
From Comment 5 in the previous review, the authors added Reference 22, which is a good reference. However, the sentence in Lines 96-98 makes no sense. The "inability" does not cause "inaccuracies". Instead, the authors might consider: "The problem of mixing DEM and SAR data is that the DEM data is often inaccurate or incomplete [22], particularly when observing waterways," making it difficult to fuse the two data types for an effective water detection technique."
[Response to Comment 2]:
Thank you very much for your professional comments. The reason you given is the true problem of the method mixing DEM and SAR data. We revised the problem using the sentence you given.
[Reviewer 1 Comment 3]:
Finally, when the reviewer mentioned that "[t]he entire scheme could use a diagram to show each of the attributes", the idea was to produce a flowchart that showed each step in the process, along with each attribute/equation that is used or affected by that step in the process. Instead, the authors just generated a table of terms in Appendix 1. While this is useful, it is not what the reviewer was anticipating. Additionally, there are so many terms to keep track of in this technique, the reviewer wonders how appealing such a complicated technique will be to other researchers.
[Response to Comment 3]:
Thank you very much for your professional comments. To show each attribute/equation that is used or affected by that step in the process, a structure diagram of all the symbols used in the water network construction is added in Appendix B. The section name of all symbols in the nomenclature of Appendix A were also added.
Yours sincerely,
Chun Liu
Aug 5, 2022
