A Self-Regulating Multi-Clutter Suppression Framework for Small Aperture HFSWR Systems

Round 1

Reviewer 1 Report

This paper proposes a novel multiple-clutter suppression framework for small aperture HFSWR systems. The proposed framework is applicable to remove sea clutter, ionosphere clutter as well as RFI. Meanwhile, it realizes tracking the travelling ship targets and finding the targets merged into clutter. Overall, the topic of this paper is interesting and timely, this work has solid contributions and offers a valuable solution to solve detection problem caused by clutter. I recommend to accept this paper after the authors address these concerns:

1. We highly recommend that authors check the grammar one more time before publication.

Specifically, in the section 1, the introduction of HFSWR has some inaccurate statements, such as “HFSWR as long-range beyond-horizon radar, has been employed for surveilling exclusive economic zone (EEZ), sea state monitoring and surveillance of aircraft and maritime vessels.”  Please revise it.

2. The introduction of previous studies is insufficient and needs to be further combed and supplemented.

3. Have you considered the computational cost of this method and whether it is efficient in online processing? You can briefly introduce it, for example, by giving relevant indicators.

4. At page 6, line 255, “The targets with low SNR (3dB-10dB) are marked as indigo blue”, But in formula (2), the green color starts with an amplitude of 0dB instead of 3dB. Is there an error between the them? Please give a proper description.

5. I understand that Figure 3 is used to show the changes of different types of clutter over a long period of time, but the target, sea clutter, ionosphere clutter and RFI cannot be clearly distinguished. Hence, in my view, Figure 3 and Figure 1 are duplicated in content.

6. The font in Figures 4, 5, and 6 seems a little small and they should be adjusted according to format requirements.

7. Why does the GSC curve in Figure 9 decline when the SNR is high? Please explain the results in detail.

Author Response

Response to Reviewer 1 Comments

Thank you very much for your detailed reviews and constructive comments, which really help us to improve the quality of the manuscript. The main corrections in the paper and the responds to the comments are as following:

Point 1: We highly recommend that authors check the grammar one more time before publication. Specifically, in the section 1, the introduction of HFSWR has some inaccurate statements, such as “HFSWR as long-range beyond-horizon radar, has been employed for surveilling exclusive economic zone (EEZ), sea state monitoring and surveillance of aircraft and maritime vessels.”  Please revise it.

Response 1: Thank you for suggestion, we have revised related section and marked them in red. In lines 26-28, the statements of “HFSWR as long-range……maritime vessels.” were corrected as “Because HFSWR provides the capability to receive target echoes over much longer distances than other traditional radars, such as microwave radars, it has been applied for surveilling exclusive economic zone (EEZ) and monitoring sea state for many years.”

Point 2: The introduction of previous studies is insufficient and needs to be further combed and supplemented.

Response 2: Thanks to you for your good comments. We have added related and lately proposed studies into our revised manuscript, please check them in section 1 and section 2.3 and 2.4. After recomposing the section 1 (introduction), the presentation of previous studies is more sufficient and well organized. We also marked all the revised content in red. In section 2.3 and 2.4, we particularly introduce the original dictionary learning (K-SVD) and compared methods (JDL and GSC).

Point 3: Have you considered the computational cost of this method and whether it is efficient in online processing? You can briefly introduce it, for example, by giving relevant indicators.

Response 3: It is really true as Reviewer suggested that the computational cost of this method is not offered in the manuscript. A part of SMSF is an online processing procedure according to each patch of received echoes, so we considered the running time of SMSF as computational cost. The offline processing procedure of SMSF is training the YOLO network, which needs about 46 minutes when epochs are set as 200. The online processing procedure of SMSF consists of training process and estimation process, training the APDL algorithm needs about 0.38 seconds and estimating current clutter echo needs 0.15 seconds. Our experiments are performed on a notebook computer configurating CPU AMD Ryzen 7 4800H, GPU NVIDIA GTX 1650 and computer memory is 16G. From the above statistics, we can confirm that SMSF has ability to suppress the received echoes data in real-time processing.

Point 4: At page 6, line 255, “The targets with low SNR (3dB-10dB) are marked as indigo blue”, But in formula (2), the green color starts with an amplitude of 0dB instead of 3dB. Is there an error between the them? Please give a proper description.

Response 4: We are very sorry for our incorrect writing, thank you for your kind suggestion, we have corrected this mistake and marked them in red. Please see the equation 8, lines 278-279, and line 289.

Point 5: I understand that Figure 3 is used to show the changes of different types of clutter over a long period of time, but the target, sea clutter, ionosphere clutter and RFI cannot be clearly distinguished. Hence, in my view, Figure 3 and Figure 1 are duplicated in content.

Response 5: As you see the target, sea clutter, ionosphere clutter and RFI cannot be clearly showed in this figure. After the deep consideration, we deleted this figure because it doesn’t contribute to observe the clutter behaviors in a long period of time.

Point 6: The font in Figures 4, 5, and 6 seems a little small and they should be adjusted according to format requirements.

Response 6: Thank you for your suggestion. We have tried our best to enlarge these figures and increase the image resolution of these figures. Please see the improvements in pages 14-17.

Point 7: Why does the GSC curve in Figure 9 decline when the SNR is high? Please explain the results in detail.

Response 7: Thank you for your valuable suggestion. It is really true as Reviewer noted that GSC curve declines when the SNR is high. This phenomenon can be explained that the null formed by auxiliary beam is shallow in a small aperture array, thereby strong target echo cannot be suppressed well. As a result, the auxiliary beam will mistakenly regard the echo signal as clutter and suppress these “clutter”, actually they are targets with strong power. Certainly, the OTDR performance of GSC will be destroyed suddenly. The related explain has been offered in the Lines 546-551 for revised manuscript.

Author Response File: Author Response.pdf

Reviewer 2 Report

Dear Authors,

   Thanks so much for your manuscript submission to MDPI Journal of Remote Sensing.  After my comprehensive evaluation, I think this paper stands for a very good set of work. The research study is both complete and convincing, and literal writing is just fairly acceptable. Hence, this research article, can be recommended as "Acceptance with Minor Edits", after fixing a few major and minor problematic issues, which I specify each of them as follows (may not limited to these as mentioned below):

   Major problematic issues suggested for improvement in your revision:
   a) Abstract: It is well organized and clearly stated. The current version is a bit long (better 180~200 words). I think the first sentence and one at Lines 15-17 can be condensed, and descriptions on your approach need to be further condensed. Also, keynote quantitative results are expected to be included in the concluding remarks.

   b) Introduction: The current version is most likely acceptable. I think the authors missed a short summary of main contributions on your research study, which can be re-edited from 2nd pargraph from the last, organized in a paragraph with 3-4 manifolds, and the right position must be placed before the last paragraphs in this section. Meanwhile, a few other minor edits are required for literal quality, i.e., hard connections such as "Thus" at corresponding Lines should be considered for replacement. 

   c) Section 2 (Materials and Methods): This section covers almost 10 pages, which looks very long. Please consider shortening the description of Algorithm 1 (SMSF framework) within a single page, and condense the other statements in each subsection. Datasets in subsection 2.6 can be shifted to the Section of Experiments and Results. Equations (2) and (5) should be re-edited in your updates. Other issues in this section are specified as below:

   d) Figures and tables: a few obvious problems must be calibrated. (i) I agree taht the authors apply the "Times New Roman"a uniform style on the characters of each figure, which should be consistent for each figure as the same font style as did for Fig. 1. (ii) Some figures have quite weak resolution in some characters (i.e., Fig. 2 and Figs. 4-7), while Fig. 9 and Fig. 11 are too small in size. The amplitude and magnitude in Y-axis can be zoomed in by 1 or 2 font size, and there are some occlusions can be removed in each of the subfigures, i.e., Figs. 4(b), 6(b) and 7(b). Please apply the enhancements as I suggested for adjusting better resolutions. (iii) Some minor issues such as distortion on the characters at X-axis, missed alignment, decent subscript on notations, can be further addressed. Please apply the required edits to make your figures and tables more professional.

   e) Discussions: the authors missed to present a section on discussing the technical innovations of their framework, the potential limitations of their approach and further improvements. If no section were arranged for this point, conducting some sensitivity analysis or ablation study are suggested for your further supplementation.

   f) Conclusions: While the conclusion section contains two passages, the statement is still a little bit generic. I think some part were expected to have been shown as partial fulfillment in the discussion section. If not, the last two sentences at the second paragraph needs some major revision. I suggest the authors expanding the last two sentences into a third paragraph with more specific details, adding a third paragraph (on prospective work or future study), with a summary of research challenges and future research orientations, etc. Please consider upgrading this crucial section to the best knowledge of your peer-reviewed authors. Thanks a lot!

   g) References: The current version looks fine, while just some minor issues should be fixed. (i) I think the authors may proceed to conduct historical review both conventional, newer and latest approaches, in one decade range (2010-2020). Also, I found that those latest keynote work within 2021-2022 are almost missing (except for [27], [28]). (ii) These latest publications in 3 Years range (2019-2022) which are similar / parallel to your study (i.e, some deep auto-encoder based dictionary learning schemes and other weakly supervised learning methods using  sparse representations), can be added in your upgraded version, which may strengthen your citations. (iii) Keep up with consistent and uniform style if citing a journal (applying abbreviated formats for each, be italic on the volume number), supplement the missed information when citing conference proceedings (the required time and location of conferences, should be updated). (iv) Check the MDPI template and see if all the requirements fulfilled, including other citations for books, online available products, etc. Thanks very much!

   Some minor issues recommended for updates in your revised version: 
   a) In addition to aligning the proper size of figures, Be sure that the actual position of each image and each table comply with the MDPI template.

   b) Literal quality of English should be further improved. There are still a few minor typos or grammatical mistakes in this version, i.e., "in the in actual operation" at Line 105, "Data set" (Dataset) at Line 368, and those hard connecting works such as "thus", "next", "and thus", etc.  I suggest the peer reviewed authors inviting a native English speaker to polish the literal aspects of this research article, including grammatical checking and careful proofreading in your updated version before resubmission.

   c) Please carefully fix any of the unprofessional notations, then check if any of the terms missed definition.  Such kind of issues need to be addressed in your proofreading procedure.

   d) If you use MS word or Latex, please avoid hyphenating a word (which currently appears multiple time at the end of some lines to cross-over two adjacent lines). MDPI online template has the options to adjust that. Thanks very much!

   Again, this paper basically stands for a good set of work, and after further editing it may qualify acceptance. Wish you the best of luck for your research article coming into acceptance. Thanks for your interests on publishing at MDPI affiliated Journals such as Remote Sensing. We expect your success in the near future. Take care!

Stay well,
With warm regards,

Author Response

Response to Reviewer 2 Comments

Thank you very much for the summary of our work and the encouraging comments. We are glad that you are satisfied with our manuscript. We tried our best to improve the manuscript and made some changes in the manuscript. These changes will not influence the content and framework of the paper. We appreciate for your warm work, and hope that the correction will meet with approval.

Point 1:  It is well organized and clearly stated. The current version is a bit long (better 180~200 words). I think the first sentence and one at Lines 15-17 can be condensed, and descriptions on your approach need to be further condensed. Also, keynote quantitative results are expected to be included in the concluding remarks.

Response 1: Thank you for your useful suggestions, we check the grammar one more time before resubmitting and rewrite the abstract. The new version is shorter than before, the first sentence and one at Lines 15-17 have been condensed. We are so sorry that the quantitative results cannot be offered because the SNR and received data are varying, which directly determine the final results, as you can find them in section 3 and section 4.

The new abstract is corrected as following

Abstract: A concerned problem in this paper is high frequency surface wave radar (HFSWR) detection of desired targets against a complex interference background consisting of sea clutter, ionosphere clutter, radio frequency interference (RFI) and atmospheric noise. Eliminating unwanted echoes and exploring obscured targets contribute to achieve ideal surveillance of sea surface targets. In this paper, a Self-regulating Multi-clutter Suppression Framework (SMSF) has been proposed for small aperture HFSWR. SMSF can remove many types of clutters or RFI, meanwhile, it mines the targets merged into clutter and tracks the travelling path of ship. In SMSF, a novel Dynamic Threshold Mapping Recognition (DTMR) method is first proposed to reduce the atmospheric noise and recognize each type of unwanted echoes, these recognized echoes are fed into proposed Adaptive Prophase-current Dictionary Learning algorithm (APDL). For making a comprehensive evaluation, we also design three novel assessment parameters, Obscured Targets Detection Rate (OTDR), Clutter Purification Rate (CPR) and Erroneous Suppression Rate (ESR). The experiment data collected from a small aperture HFSWR system confirms that SMSF has a precise suppression performance over most of the classical algorithms and concurrently reveals the moving targets, and OTDR of SMSF is usually higher than compared methods.

Point 2: Introduction: The current version is most likely acceptable. I think the authors missed a short summary of main contributions on your research study, which can be re-edited from 2nd pargraph from the last, organized in a paragraph with 3-4 manifolds, and the right position must be placed before the last paragraphs in this section. Meanwhile, a few other minor edits are required for literal quality, i.e., hard connections such as "Thus" at corresponding Lines should be considered for replacement.

Response 2: We appreciate your valuable suggestions. We have corrected the hard connections and improved the expressions in introduction. All the improvements are marked in red. We also added the summary of main contributions on our research in Lines 113 -125 and marked them in red. Please see as following.

The main contributions are described as follows: The first one is that designing a novel denoising process DTMR in the target detection. It realizes decontaminating RD images and recognizing each type of unwanted echoes. By incorporating the deep learning and semi-empirical model to preciously classify unwanted echoes components, the data collection is more efficient and reliable. The second one is proposing an adaptive dictionary learning algorithm, which is a self-regulating dictionary according to current echoes. Certainly, it will fit the varying clutter in various conditions. The third one is SMSF framework achieves suppressing many types of clutter simultaneously, which is suitable observing targets in a long period of time. The fourth one is proposed assessment parameters, Obscured Targets Detection Rate (OTDR), Clutter Purification Rate (CPR) and Erroneous Suppression Rate (ESR). This is the first time to access suppression results in quantitative indicators and it also can be viewed as a more global approach to evaluate SMSF and other classical suppression methods.”

Point 3: Section 2 (Materials and Methods): This section covers almost 10 pages, which looks very long. Please consider shortening the description of Algorithm 1 (SMSF framework) within a single page, and condense the other statements in each subsection. Datasets in subsection 2.6 can be shifted to the Section of Experiments and Results. Equations (2) and (5) should be re-edited in your updates. Other issues in this section are specified as below:

Response 3: Thank you for your suggestion, we have shorten the length of Section 2 (Materials and Methods), and tried our best to shorten the description of Algorithm 1. The subsection 2.6 has been shifted to the section of Results, please see Page 12. But reviewer 3 suggested us to add introduction of compared methods, the entire length of section 2 is not shorter than before. The content of section 2 become more breadth and balance.

Point 4: Figures and tables: a few obvious problems must be calibrated. (i) I agree taht the authors apply the "Times New Roman"a uniform style on the characters of each figure, which should be consistent for each figure as the same font style as did for Fig. 1. (ii) Some figures have quite weak resolution in some characters (i.e., Fig. 2 and Figs. 4-7), while Fig. 9 and Fig. 11 are too small in size. The amplitude and magnitude in Y-axis can be zoomed in by 1 or 2 font size, and there are some occlusions can be removed in each of the subfigures, i.e., Figs. 4(b), 6(b) and 7(b). Please apply the enhancements as I suggested for adjusting better resolutions. (iii) Some minor issues such as distortion on the characters at X-axis, missed alignment, decent subscript on notations, can be further addressed. Please apply the required edits to make your figures and tables more professional.

Response 4: We are very sorry for our negligence, thank you for kind and scrupulous remind. We have updated all the figures in this manuscript according to each comment. We adjust the image resolution of figure 4-7 and figure 2, and also enlarge the figure 9 and figure 11. Moreover, the characters at X and Y axis are adjusted. Please see the line 246, page 14-17.

Point 5:  Discussions: the authors missed to present a section on discussing the technical innovations of their framework, the potential limitations of their approach and further improvements. If no section were arranged for this point, conducting some sensitivity analysis or ablation study are suggested for your further supplementation.

Response 5: According to your comment, we have added a section (Section 4.4) for discussing the technical innovations of proposed. We marked this section in red, you can find them in lines 585-604. And this section is introduced as following.

All the experiments results have approved that SMSF have expected suppression performance and finding target ability, which attributes to the technical innovations of SMSF. The foremost innovations are strong adaptability of SMSF for varying echoes data, and self-regulating ability of SMSF in real-time processing. SMSF can express the unwanted echoes in sparse representation by adequately learning historical data to find their similarities. And SMSF learns the current data in good time for capturing latest information. Under this design, SMSF reduces the sensitivity of data selection and also decreases the dependence on training set, which the second innovations. According to the latest received clutter data as a reference, the information learned by historical data and the current data are optimized in real time, the balance between the two different learning stages is adjusted to fit the latest received clutter data, thus the optimal representation can be found. SMSF as a remedy to solve the problem of JDL, the final suppression results are not limited by training sample selection. The third innovation of SMSF is its wide applicability, SMSF can suppress many types of clutter and RFI. By incorporating deep learning network, every type of clutter is preciously sorted out for dictionary learning. This design surpasses a perennial obstacle that each type of clutter is suppressed by each specific method. However, SMSF doesn't use spatial information to their full potential, and the suppression performance is not ideal when the SCR (Signal Clutter Ratio) is low, so the dimension of dictionary learning is further increased in the future work.

Point 6:  Conclusions: While the conclusion section contains two passages, the statement is still a little bit generic. I think some part were expected to have been shown as partial fulfillment in the discussion section. If not, the last two sentences at the second paragraph needs some major revision. I suggest the authors expanding the last two sentences into a third paragraph with more specific details, adding a third paragraph (on prospective work or future study), with a summary of research challenges and future research orientations, etc. Please consider upgrading this crucial section to the best knowledge of your peer-reviewed authors. Thanks a lot!

Response 6: Thank you for your precious suggestion, we have made adjustment according to the Reviewer’s comments. We expanded the last two sentences into a third paragraph with more specific research plan in the future, which is the current conception to achieve a series of tasks, clutter suppression, target detection and target tracking. Please refer to the lines 627-633.

Point 7: References: The current version looks fine, while just some minor issues should be fixed. (i) I think the authors may proceed to conduct historical review both conventional, newer and latest approaches, in one decade range (2010-2020). Also, I found that those latest keynote work within 2021-2022 are almost missing (except for [27], [28]). (ii) These latest publications in 3 Years range (2019-2022) which are similar / parallel to your study (i.e, some deep auto-encoder based dictionary learning schemes and other weakly supervised learning methods using sparse representations), can be added in your upgraded version, which may strengthen your citations. (iii) Keep up with consistent and uniform style if citing a journal (applying abbreviated formats for each, be italic on the volume number), supplement the missed information when citing conference proceedings (the required time and location of conferences, should be updated). (iv) Check the MDPI template and see if all the requirements fulfilled, including other citations for books, online available products, etc. Thanks very much!

Response 7: Thank you very much for reminding me and these intensive comments are very useful for improving our manuscript. We conduct a historical review on both conventional and newer approaches in one decade range (2010-2020). Moreover, we also added some latest studies into the introduction. As Reviewer suggested that deep auto-encoder based dictionary learning and other methods using sparse representations are also added into this revised manuscript. Please see the new reference in pages 22-23, and we list them the following.

1. Wang, Z. Q.; Li, Y. J.; Shi, J. N.; Wang, P. F.; Chen, D. H. Spread Sea Clutter Suppression in HF Hybrid Sky-Surface Wave Radars Based on General Parameterized Time-Frequency Analysis. Int J Antenn Propag. 2020, 3:1-12.
2. Doulamis, A. D.; Shang, S.; He, K.N.; Wang, Z.B.; Yang, T.; Liu, M.; Li, X. Sea Clutter Suppression Method of HFSWR Based on RBF Neural Network Model Optimized by Improved GWO Algorithm. Computational Intelligence and Neuroscience. 2020, https://doi.org/10.1155/2020/8842390.
3. Hua, X.; Ono, Y.; Peng, L.; Cheng, Y.; Wang, H. Target detection within nonhomogeneous clutter via total bregman divergence-based matrix information geometry detectors. IEEE Trans. Signal Process. 2021, 69, 4326-4340.
4. Zhang, L.; Li, Q.F. Wu, Q. M. Jonathan. Target Detection for HFSWR Based on an S³D Algorithm. IEEE Access, 2020, 8, 224825-224836.
5. Azhagu Jaisudhan Pazhani, A.; Vasanthanayaki, Object detection in satellite images by faster R-CNN incorporated with enhanced ROI pooling (FrRNet-ERoI) framework. ESIN. 2022, 15, 553–561.
6. Liu, D.Y.; Liang, C.W.; Chen, S.K.;  Tie, Y.; Qi, L. Auto-encoder based structured dictionary learning for visual classification, Neurocomputing. 2021, 438, 34-43.
7. Wang, Z.; Shi, S.; He, Z.; Sun, Guo.; Cao, J. An ocean clutter suppression method for OTHR by combining optimal filter and dictionary learning. 2018 IEEE Radar Conference (RadarConf18). Oklahoma City, USA, April 2018, 1499-1503.
8. Zhang, X.; Yao, D.; Yang, Q.; Dong, Y.N.; Deng, W.B. Knowledge-Based Generalized Side-Lobe Canceller for Ionospheric Clutter Suppression in HFSWR. Remote Sens.2018, 10, 104.

Some minor issues recommended for updates in your revised version: 
Point 8:  In addition to aligning the proper size of figures, Be sure that the actual position of each image and each table comply with the MDPI template.

Response 8: We are very sorry for our negligence of this problem, we have adjusted the position of each image and each table comply with the MDPI template. Thank you for reminding me.

Point 9:  Literal quality of English should be further improved. There are still a few minor typos or grammatical mistakes in this version, i.e., "in the in actual operation" at Line 105, "Data set" (Dataset) at Line 368, and those hard connecting works such as "thus", "next", "and thus", etc.  I suggest the peer reviewed authors inviting a native English speaker to polish the literal aspects of this research article, including grammatical checking and careful proofreading in your updated version before resubmission.

Response 9: We are very sorry for our incorrect writing, we have corrected these mistakes please see line 106 and line 392, and we improved the manuscript including grammatical checking and careful proofreading.

Point 10: Please carefully fix any of the unprofessional notations, then check if any of the terms missed definition.  Such kind of issues need to be addressed in your proofreading procedure.

Response 10: We have made correction according to the Reviewer’s comments and proofread the procedure.

Point 11: If you use MS word or Latex, please avoid hyphenating a word (which currently appears multiple time at the end of some lines to cross-over two adjacent lines). MDPI online template has the options to adjust that. Thanks very much!

Response 11: Special thanks to you for your good comments. We have made some adjustments on this problem.

At last, we would like to thank you again for your valuable comments, which have greatly helped us to improve the content, quality, and presentation of this paper. We hope you are satisfied with our revisions. If there are further comments, we will be more than happy to address them as well.

Author Response File: Author Response.pdf

Reviewer 3 Report

The authors analysed the high frequency surface wave radar (HFSWR) detection of desired targets against a complex interference background consisting of sea clutter, ionosphere clutter, radio frequency interference (RFI) and background noise. In general, the paper is consistent and justified, but the language must be improved. The major problem of this manuscript is that the previous related work is not explained, and, consequently, the results obtained are not compared with other results. Thus, the presentation of related work is needed in the introduction, and the discussion section must be added.

Author Response

Response to Reviewer 3 Comments

Thank you so much for allowing a revision of our manuscript, with an opportunity to address the comments. We find these comments are very constructive and helpful. We have tried our best to demonstrate our explanations and revisions in response to the following comments.

Point 1: The authors analysed the high frequency surface wave radar (HFSWR) detection of desired targets against a complex interference background consisting of sea clutter, ionosphere clutter, radio frequency interference (RFI) and background noise. In general, the paper is consistent and justified, but the language must be improved. The major problem of this manuscript is that the previous related work is not explained, and, consequently, the results obtained are not compared with other results. Thus, the presentation of related work is needed in the introduction, and the discussion section must be added.

Response 1: Special thanks to you for your good comments.

Considering the Reviewer’s suggestion, we have made correction.

1. First, we improved expression in this revised manuscript including grammatical checking and proofreading before resubmission.
2. Second, we have added the introduction of previous related work in section 2, which includes the K-SVD algorithm and two compared methods (GSC and JDL). You can see added content in lines 186-232 and we marked related content in red.
3. Third, we also carefully compare the performance among the APDL, ADL and PDL in Section 3. As for the suppressed results comparison of ionosphere clutter, please see the lines 451-458, lines 468-472 in Section 3.4. As for the suppressed results comparison of sea clutter, please see the lines 485- 486 and lines 491-495 in Section 3.5. The suppression results comparison of RFI are also improved in Section 3.6.
4. Fourth, we added the Section 4.4 (Technical Innovations of SMSF) to analyze the distinction of SMSF for classical compared methods (JDL and GSC). And we also improved the Section 4 to fully discuss the detection ability and suppression performance for each type of approach. All the revised content have been marked in red.
5. The Section 4.4 as following.

   4.4 Technical Innovations of SMSF

All the experiments results have approved that SMSF have expected suppression performance and finding target ability, which attributes to the technical innovations of SMSF. The foremost innovations are strong adaptability of SMSF for varying echoes data, and self-regulating ability of SMSF in real-time processing. SMSF can express the unwanted echoes in sparse representation by adequately learning historical data to find their similarities. And SMSF learns the current data in good time for capturing latest information. Under this design, SMSF reduces the sensitivity of data selection and also decreases the dependence on training set, which the second innovations. According to the latest received clutter data as a reference, the information learned by historical data and the current data are optimized in real time, the balance between the two different learning stages is adjusted to fit the latest received clutter data, thus the optimal representation can be found. SMSF as a remedy to solve the problem of JDL, the final suppression results are not limited by training sample selection. The third innovation of SMSF is its wide applicability, SMSF can suppress many types of clutter and RFI. By incorporating deep learning network, every type of clutter is preciously sorted out for dictionary learning. This design surpasses a perennial obstacle that each type of clutter is suppressed by each specific method. However, SMSF doesn't use spatial information to their full potential, and the suppression performance is not ideal when the SCR (Signal Clutter Ratio) is low, so the dimension of dictionary learning is further increased in the future work.

At last, we would like to thank you again for your valuable comments, which have greatly helped us to improve the content, quality, and presentation of this paper. We hope you are satisfied with our revisions. If there are further comments, we will be more than happy to address them as well.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors have addressed all my concerns.

Reviewer 3 Report

The authors revised the manuscript according to the previous comments.