A Novel Edge Detection Method for Multi-Temporal PolSAR Images Based on the SIRV Model and a SDAN-Based 3D Gaussian-like Kernel

Round 1

Reviewer 1 Report

In this paper, the authors proposed a novel edge detection method for multi-temporal PolSAR images based on the SIRV model and a SDAN-based 3-D Gaussian-like kernel. Two real multi-temporal PolSAR datasets are used for the comparison and performance evaluation. Overall, this paper is well written with enough contributions and experiments. I have some minor comments as follows:

1)     It would be better to give quantitative results of the proposed method over other state-of-the-art methods.

2)     For the PolSAR datasets, the azimuth and range directions should be indicated in the Pauli-coded images, such as the figure 1 and figure 6. Please have a check.

3)     What is the advantage of the SIRV model used for the PolSAR edge detection?

4)     How to obtain the ground truth of the edge map? The authors should explain this issue.

5)     The disadvantages of the proposed method in comparison with the state-of-the-art methods should be discussed, such as the computational complexity.

Author Response

Thank you very much for your comments concerning our manuscript entitled “A Novel Edge Detection Method for Multi-temporal PolSAR Images Based on the SIRV Model and a SDAN-based 3-D Gaussian-like Kernel”. Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our research. We have studied the comments carefully and have made corrections that we hope to meet with approval. Corresponding changes have been made to the manuscript with red color. Responses to the reviewers’ comments are presented as follows.

 

Replies to the Comments of Reviewer #1

In this paper, the authors proposed a novel edge detection method for multi-temporal PolSAR images based on the SIRV model and a SDAN-based 3-D Gaussian-like kernel. Two real multi-temporal PolSAR datasets are used for the comparison and performance evaluation. Overall, this paper is well written with enough contributions and experiments. I have some minor comments as follows:

 

1. It would be better to give quantitative results of the proposed method over other state-of-the-art methods.

• Dear reviewer, many thanks for your comment. At present, there are few studies for multi-temporal PolSAR image edge detection. In this paper, three sets of comparison experiments are conducted, namely 3-D mean kernel, 3-D maximum kernel and 3-D RMS kernel. In the revised manuscript, we have given the quantitative results of comparison methods in Table 2 and Table 3.

 

2. For the PolSAR datasets, the azimuth and range directions should be indicated in the Pauli-coded images, such as the figure 1 and figure 6. Please have a check.

• Many thanks for your comment. In the PolSAR datasets provided by the AgriSAR 2009 Campaign project, the project gives the azimuth of the PolSAR datasets. In addition, an important note that we missed in the original manuscript is that the each multi-temporal PolSAR dataset was acquired on the same satellite orbit. In the revised manuscript, we have added the dataset description.

 

3. What is the advantage of the SIRV model used for the PolSAR edge detection?

• Many thanks for your comment. The SIRV model overcomes the problem of heterogeneous regions not satisfying Gaussian clutter modelling, and its application to edge detection can better determine whether pixels in heterogeneous regions are at edge locations. In the revised manuscript, we have explained the advantages of the SIRV model.

 

4. How to obtain the ground truth of the edge map? The authors should explain this issue.

• Many thanks for your comment. The ground truths of the edge maps are obtained from the corresponding multi-temporal optical images. In the revised manuscript, we have explained the origin of the ground truth edges.

 

5. The disadvantages of the proposed method in comparison with the state-of-the-art methods should be discussed, such as the computational complexity.

• Many thanks for your comment. In the revised manuscript, we have provided a detailed comparative analysis of the experimental results. In particular, a note on the computational complexity has been added.

Reviewer 2 Report

Minor errors like the definite article "the" in row 85 of the paper.

The article would be improved if there is a brief description how the polarimetric SAR images are formed (or which polarizations are used, for example "HH + HV" etc.) from multiple polarization Radarsat-2 data.

Author Response

Thank you very much for your comments concerning our manuscript entitled “A Novel Edge Detection Method for Multi-temporal PolSAR Images Based on the SIRV Model and a SDAN-based 3-D Gaussian-like Kernel”. Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our research. We have studied the comments carefully and have made corrections that we hope to meet with approval. Corresponding changes have been made to the manuscript with red color. Responses to the reviewers’ comments are presented as follows.

Replies to the Comments of Reviewer #2

This manuscript proposes a Siamese Dense Capsule Network (SD-CapsNet) for SAR image registration. A texture constraint-based phase congruency is proposed to detect uniformly distributed keypoints of high repeatability. SD-CapsNet is used to implement feature descriptor extraction and matching which can get better semantic information. In this manuscript, the content is detailed and the experimental data are sufficient. I have some comments and suggestions.

1. Minor errors like the definite article "the" in row 85 of the paper.

• Dear reviewer, many thanks for your comment. We are really sorry for our careless mistakes, in the revised manuscript, we have corrected those errors.

 

2. The article would be improved if there is a brief description how the polarimetric SAR images are formed (or which polarizations are used, for example "HH + HV" etc.) from multiple polarization Radarsat-2 data.

• Many thanks for your comment. In this paper, we used two quad-polarimetric SAR datasets for evaluation, which consist of HH, HV, VH, VV channels. In the revised manuscript, we have given the specific polarizations of the PolSAR images.

 

Reviewer 3 Report

This manuscript proposes an optimized process to detect edges for multi-temporal PolSAR images, which is composed of SIRV model, SDAN, and 1-D convolution kernel. The manuscript validates the algorithm in PolSAR data for evaluation and analyzes the impact of the algorithm.

Some problems of this manuscript are listed below:

1.     Is the inference of chi-square distribution based on the data distribution of the POLSAR data? In addition, whether the chi-square distribution fits the probability density perfectly?

2.     Regarding the derivation of formula (22), some hyperparameters such as 0.4, 0.02 should be given specific meaning.

3.     The visual analysis of Fig.7 should be improved. If the ground truth of edge is given, Fig.7 will be more intuitive.

4.     This paper attribute advantages of the proposed method to differences in scattering mechanisms, is this difference manageable? Will there be a worse effect after 1-D convolution?

5.     In Fig.8, what are the meanings of the green and red parts?

6.     There are some grammar mistakes, such as the usage of articles and transition words. It is better to correct them all.

In conclusion, this manuscript elaborated theoretical basis of the relevant methodology, but there is less analysis of the proposed method and underlying reasons, major revision should be carried out. 

Author Response

Thank you very much for your comments concerning our manuscript entitled “A Novel Edge Detection Method for Multi-temporal PolSAR Images Based on the SIRV Model and a SDAN-based 3-D Gaussian-like Kernel”. Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our research. We have studied the comments carefully and have made corrections that we hope to meet with approval. Corresponding changes have been made to the manuscript with red color. Responses to the reviewers’ comments are presented as follows.

Replies to the Comments of Reviewer #3

This manuscript proposes an optimized process to detect edges for multi-temporal PolSAR images, which is composed of SIRV model, SDAN, and 1-D convolution kernel. The manuscript validates the algorithm in PolSAR data for evaluation and analyzes the impact of the algorithm.

Some problems of this manuscript are listed below:

 

1. Is the inference of chi-square distribution based on the data distribution of the POLSAR data? In addition, whether the chi-square distribution fits the probability density perfectly?

• Dear reviewer, many thanks for your comment. The inference of the chi-square distribution was not obtained on the basis of PolSAR data, it is obtained from the shape of the probability density function. We notice that the number of edge positions is much less than that of non-edge positions in most scenes, and the probability density of non-edges is much greater than that of edges. Therefore, the probability density function can basically be fitted to a chi-square distribution. In the manuscript, we have explained the reasons for choosing the chi-square distribution to fit the probability density function.

2. Regarding the derivation of formula (22), some hyperparameters such as 0.4, 0.02 should be given specific meaning.

• Many thanks for your comment. In the revised manuscript, we have added some references to give the basis for the choice of hyperparameters.

3. The visual analysis of Fig.7 should be improved. If the ground truth of edge is given, Fig.7 will be more intuitive.

• Many thanks for your suggestion. In the original manuscript, the ground truth of edges is given in Fig.6. According to your suggestion, in the revised manuscript, we have added the ground truth edge map in Fig.7.

4. This paper attribute advantages of the proposed method to differences in scattering mechanisms, is this difference manageable? Will there be a worse effect after 1-D convolution?

• Many thanks for your comment. In fact, for land covers with time-varying scattering mechanisms (e.g. vegetation and crops), different land covers may exhibit similar scattering mechanisms in a single-date PolSAR image, making it difficult to extract the edges of these land covers. Differences in scattering mechanisms between different land covers at different dates are unmanageable. Our proposed 1-D convolutional kernel is used to smooth out edge strength variations in the temporal dimension and reduce false edge pixels caused by speckle noise. In the revised manuscript, we have explained this comment.

5. In Fig.8, what are the meanings of the green and red parts?

• Many thanks for your comment. In Fig.8, both the red and green boxes indicate regions of incomplete edge detection in single-date PolSAR image. In the following, we have only analyzed the region marked by green box, therefore we use two colors to distinguish the different regions. In the revised manuscript, we have explained the meanings of the green and red boxes.

6. There are some grammar mistakes, such as the usage of articles and transition words. It is better to correct them all.

• Many thanks for your comment. In the revised manuscript, we have checked and corrected some grammar errors.

In conclusion, this manuscript elaborated theoretical basis of the relevant methodology, but there is less analysis of the proposed method and underlying reasons, major revision should be carried out.

• We tried our best to improve the manuscript and made some changes in the manuscript. We appreciate for Editors/Reviewers’ warm work earnestly, and hope that the correction will meet with approval. Once again, thank you very much for your comments and suggestions.

Round 2

Reviewer 3 Report

In the newly submitted version of the paper, all suggestions have been replied to and addressed. And a theoretical question needs to be solved. In the revised version, the authors emphasize the conclusion about chi-square distributions are derived from observations. The probability density function (PDF) of ESM approximately satisfies the chi-square distribution. In fact, such a hypothesis is not convincing, if quantitative criteria cannot be given.

Author Response

Dear Reviewer,

Thanks for your comments and suggestion. We are sorry for the unclear descriptions about the PDF of ESM. In the revised version, we corrected this statement.