A Fast Hyperspectral Anomaly Detection Algorithm Based on Greedy Bilateral Smoothing and Extended Multi-Attribute Profile
Round 1
Reviewer 1 Report
The manuscript is well written. This reviewer only provide comments for improvements:
Line 43, should 'Reed-Xiaoli' be 'Reed-Yu' (based on the cited paper's author names)? I mean it may be better to use authors last names to refer the method.
Lines 39 and 40, 'recent' is used twice. One of them can be changed to a different word with the same meaning.
Line 94, 'EMAP' was used in many places within the manuscript. Since there is a Germany 'EnMap' satellite instrument. I am not sure that if the authors of the present manuscript should use a different notation from 'EMAP' to avoid possible confusions with 'EnMap'.
Lines 100 - 108 should be deleted or be added to Section 5 'Conclusions'.
Line 204, 'designated' can be changed to 'designated as' or 'designated by'.
Line 247, I think that both 'nm' should be removed. The numbers 144-153 and 196-107 should be band numbers. AVIRIS does not cover 144-153 nm spectral range. Also, '107' should most likely be '207'.
Line 266, 'China' should be changed to 'China'.
Finally, I would like to provide a more general comment regarding the selected 5 scenes. All the anomaly features in these cases have obvious spatial features. In many situations, the anomaly pixels do not have obvious spatial features. I am not sure if the proposed method by the quthors would really work in these situations. For example, the US Curprite site in Nevada has been widely used for testing geological remote sensing algorithms. The AVIRIS Cuprite data sets acquired over different years are most likely publicly available. Drs. Roger Clark and G. Swaze were able to map over 100 types of minerals from AVIRIS Cuprite data sets. I am not sure if the authors' method would be able to detect pixels containing special mineral materials from AVIRIS data sets. After all, the original concept of imaging spectrometry proposed by Drs. A. Goetz and G. Vane was for geological remote sensing.
Author Response
Response to Reviewer 1
Thank you for your careful reading of the manuscript and constructive remarks concerning our manuscript entitled “A Fast Hyperspectral Anomaly Detection Algorithm Based on Greedy Bilateral Smoothing and Extended Multi-Attribute Profile” (remotesensing-1400066). Those comments are all valuable and very helpful for revising and improving our paper and the important guiding significance to our research. We have studied comments carefully and have made corrections which we hope meet with approval. Revised portions are used “Track changes” function in Microsoft Word. The main corrections in the paper and the responses to the reviewers’ comments are as follows: (reviewer’s comments in black, our replies in red. The index of the number of rows mentioned in the reply is in the "Track changes-All Markup" mode.):
Point 1: Line 43, should 'Reed-Xiaoli' be 'Reed-Yu' (based on the cited paper's author names)? I mean it may be better to use authors last names to refer the method.
Response 1: Thanks for your suggestion. The name of the algorithm 'RX' is a conventional name. It was not proposed by the author but was named by E. A. Ashton and A. Schaum in 1998 when citing the algorithm article: "A new sub-pixel target detection algorithm is developed that integrates a linear mixing model (LMM) with the powerful "RX" anomaly detector of Reed and Yu (1990). [1]". This may be a minor mistake, but the RX algorithm and Its name 'RX' are widely used. So most scholars use 'Reed-Xiaoli' to introduce the RX algorithm they cited. In addition, many RX improved algorithms also use 'RX' as part of the algorithm name, such as KRX, SSRX, and so on. Therefore, this article uses 'Reed-Xiaoli' as the full name to introduce the algorithm.
- Ashton, E.A.; Schaum, A. Algorithms for the detection of sub-pixel targets in multispectral imagery. PHOTOGRAMMETRIC ENGINEERING AND REMOTE SENSING 1998, 64, 723-731.
Point 2: Lines 39 and 40, 'recent' is used twice. One of them can be changed to a different word with the same meaning.
Response 2: Thanks for your suggestion. We have changed 'In recent decades' in line 40 (All Markup) to 'In the past few decades'.
Point 3: Line 94, 'EMAP' was used in many places within the manuscript. Since there is a Germany 'EnMap' satellite instrument. I am not sure that if the authors of the present manuscript should use a different notation from 'EMAP' to avoid possible confusions with 'EnMap'.
Response 3: Thanks for your suggestion. We have added words and subscripts such as 'algorithm', 'method', 'features' after all 'EMAP' to ensure that it is used as a method or extracted features to avoid confusion. Please see lines 164, 170, and 208 (All Markup). In addition, we reviewed the relevant literature and confirmed the feasibility of using this abbreviation.
Point 4: Lines 100 - 108 should be deleted or be added to Section 5 'Conclusions'.
Response 4: Thanks for your suggestion. We have deleted the part about the algorithm contribution from 100-108 lines.
Point 5: Line 204, 'designated' can be changed to 'designated as' or 'designated by'.
Response 5: Thanks for your suggestion. We have changed 'designated' to 'designated as'.
Point 6: Line 247, I think that both 'nm' should be removed. The numbers 144-153 and 196-107 should be band numbers. AVIRIS does not cover 144-153 nm spectral range. Also, '107' should most likely be '207'.
Response 6: Thank you so much for your careful check, and we feel sorry for the inconvenience brought to the reviewer. We have removed "nm" and corrected ‘107’ to ‘207’. At the same time, we checked other parts of the manuscript to ensure that there were no similar errors.
Point 7: Finally, I would like to provide a more general comment regarding the selected 5 scenes. All the anomaly features in these cases have obvious spatial features. In many situations, the anomaly pixels do not have obvious spatial features. I am not sure if the proposed method by the quthors would really work in these situations. For example, the US Curprite site in Nevada has been widely used for testing geological remote sensing algorithms. The AVIRIS Cuprite data sets acquired over different years are most likely publicly available. Drs. Roger Clark and G. Swaze were able to map over 100 types of minerals from AVIRIS Cuprite data sets. I am not sure if the authors' method would be able to detect pixels containing special mineral materials from AVIRIS data sets. After all, the original concept of imaging spectrometry proposed by Drs. A. Goetz and G. Vane was for geological remote sensing.
Response 7:
Thanks for your suggestion. In the application scenario of processing sub-pixels, the function of GBSAED is indeed subject to some restrictions. According to your question about the applicability of the algorithm, we have made the following supplements in the Conclusion, please see lines 500-502 (All Markup): “GBSAED algorithm has a better performance in processing anomalies with spatial characteristics rather than dealing with subpixels. Therefore, our further direction will focus on the adaptability improvement of the algorithm.”
For the mineral mapping problem you mentioned, our algorithm can identify some rare minerals exposed on the surface because, compared to ordinary rocks, these minerals occupy a smaller number of pixels in the image and have special spectral characteristics. However, because hyperspectral anomaly detection is essentially a binary classification problem, we can only detect these special minerals as high-value pixels but cannot determine their category. We will study this work later in the research and development of supervised hyperspectral target detection algorithms. By introducing sample vectors, specific types of minerals will be more accurately identified.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Dear authors,
After reviewing your article I found it interesting and I think it could be interesting to the Journal's readers. Before I could recommend to be considered for publication, minor to moderate changes are necessary. Please find the details in the attached documents.
Kind regards,
Reviewer
Comments for author File: 
Comments.pdf
Author Response
Response to Reviewer 2
Thank you for your careful reading of the manuscript and constructive remarks concerning our manuscript entitled “A Fast Hyperspectral Anomaly Detection Algorithm Based on Greedy Bilateral Smoothing and Extended Multi-Attribute Profile” (remotesensing-1400066). Those comments are all valuable and very helpful for revising and improving our paper and the important guiding significance to our research. We have studied comments carefully and have made corrections which we hope meet with approval. Revised portions are used “Track changes” function in Microsoft Word. The main corrections in the paper and the responses to the reviewers’ comments are as follows: (reviewer’s comments in black, our replies in red. The index of the number of rows mentioned in the reply is in the "Track changes-All Markup" mode.):
Point 1: Section 1. Introduction
The Introduction section is extensive. In my opinion, if applicable and not necessary, it could be shortened a bit to increase readability. More precisely, from line 40 to line 89. The information here is important, however, if there is a way to present it a more concise manner. Details on the noted algorithms could be moved to the second section, and only mention them in the Introduction section.
Response 1:
Special thanks to you for your comments. According to your suggestion, we have condensed the introduction, mainly including the RX algorithm and the algorithm introduction based on the expression class. Please see lines 44-51 and 60-77. After the correction, we only introduced the simple principle of RX because it is the benchmark for anomaly detection algorithms. The principles and methods of the remaining specific algorithms have all been deleted, and only the overall ideas of the algorithm categories are introduced.
The overall framework of the introduction part is as follows. We first introduce the deficiencies of the benchmark RX for anomaly detection and its improved methods. Secondly, we present representation-based algorithms. Then introduce the advantages of the LRaSMD method in hyperspectral anomaly detection. Finally, based on the deficiencies of the LRaSMD method, we propose our method.
Point 2: Section 2. Related work
Addresses low-rank and sparse matrix decomposition and extended morphological attribute profile.
This section could be extended from the Introduction section. This would reduced the length of the Introduction and increase the length of the Related work section.
Response 2:
Thank you very much for your suggestion. Since the Related work section mainly introduces the underlying principles and algorithm formulas of the predecessor of the proposed algorithm, the content of the algorithm principles in the Introduction and the proposed algorithm does not belong to the same category. Therefore, it was only streamlined in the Introduction and not added to the Related work. We gratefully appreciate your valuable suggestion.
Point 3: Section 3. Proposed method
Appropriate detail. However, please note and revise: Every Figure/Graph/Table/Equation has to be mentioned in the text (paragraph) close to it. Further, every Figure/Graph/Table/Equation has to be briefly discussed or analyzed.
Contains all necessary information.
There should be no the text between titles and subtitles.
There should be no text between 3. and 3.1.
Instead, introduce an additional subtitle.
- Proposed method
3.1. GBSAED method flowchart
3.2. Fast extraction of abnormal spectral features using greedy bilateral smoothing
3.3. Extracting abnormal spatial features based on the extended multi-attribute profile
3.4. Proposed GBSAED algorithm
Response 3:
Thanks for your suggestion. We have ensured that there is no text between the title and subtitles and have revised each title as follows, please see lines 166, 167, 175, 206, 218:
- Proposed method
3.1. GBSAED method flowchart
3.2. Fast extraction of abnormal spectral features using greedy bilateral smoothing
3.3. Extracting abnormal spatial features based on the extended multi-attribute profile
3.4. Proposed GBSAED algorithm
Point 4: Section 4. Experimental results and analysis
Provides sufficient detail. However, please note and revise: Every Figure/Graph/Table/Equation has to be mentioned in the text (paragraph) close to it. Further, every Figure/Graph/Table/Equation has to be briefly discussed or analyzed.
Contains all necessary information.
There should be no the text between titles and subtitles.
There should be no text between 4. and 4.1.
Instead, introduce an additional subtitle.
- Experimental results and analysis
4.1. Experiment setup (or something similar)
4.2. Hyperspectral datasets
4.3. Detection performance
4.4. Parameter setting considerations
Response 4:
Thanks for your suggestion. We have ensured that there is no text between the title and subtitles and have revised each title as follows, please see lines 244, 245, 253, 290, 433:
- Experimental results and analysis
4.1. Experiment setup
4.2. Hyperspectral datasets
4.3. Detection performance
4.4. Parameter setting considerations
In addition, based on the suggestions you mentioned in the conclusion section to supplement the discussion of the latest algorithms, we added two algorithms and conducted experiments on five data sets. The first algorithm is Graph and Total Variation Regularized Low-Rank Representation based detector (GTVLRR)[1], which is the same type of algorithm as the LRASR used in the original manuscript. The former is an improved method of the latter, so we use GTVLRR instead of the original LRASR algorithm. The second algorithm is the parts representation-based low rank and sparse matrix decomposition anomaly detector (PRLRaSAD)[2], PRLRaSAD and the proposed algorithm belong to the LRaSMD category, so this algorithm is selected for discussion. The adjustments to this part are summarized as follows:
(1) Use these two algorithms to conduct experiments on five data sets and evaluate them under each evaluation index. Please see lines 321, 322.
(2) Both algorithms were tuned. GTVLRR was tuned with reference to the parameters suggested by the author's paper. PRLRaSAD was tuned after consulting and discussing with the original author. Please see line 313.
(3) Re-drawn all figures in the Experimental results and analysis section. Please see lines 338, 343, 360, 365, 380, 385, 402, 407, 425, 430.
(4) A comparative analysis of these two algorithms is added in the result discussion section.
- Cheng, T.; Wang, B. Graph and Total Variation Regularized Low-Rank Representation for Hyperspectral Anomaly Detection. IEEE Transactions on Geoscience and Remote Sensing 2020, 58, 391-406, doi:10.1109/TGRS.2019.2936609.
- Zhang, Y.; Fan, Y.; Xu, M.; Li, W.; Zhang, G.; Liu, L.; Yu, D. An Improved Low Rank and Sparse Matrix Decomposition-Based Anomaly Target Detection Algorithm for Hyperspectral Imagery. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 2020, PP, 1-1.
Point 5: Section 5. Conclusions
Please address and discuss additional studies in this domain.
Preferably newer studies from 2020 and 2021 if applicable.
Compare and comment the results from other studies and methods, compared to your approach.
Discuss the contribution of the manuscript and its potential applications and implications.
Response 5:
Thank you very much for your suggestion. We have added two sets of experiments in the Experimental results and analysis section, which are algorithms published in 2020. In Response 4, we introduced our supplementary work. In the conclusion section, we discussed the detection performance, operating efficiency, and robustness of the algorithm.
Point 6: Section 6. Conclusion
Please address the limitations and advantages of your manuscript. Briefly describe approximate guidelines and ideas for future research should be added.
Highlight the significance of the study and the approach to the existing body of literature. Address practical implications.
Response 6:
Thanks for your suggestion. In the application scenario of processing sub-pixels, the function of GBSAED is indeed subject to some restrictions. According to your question about the limitations of the algorithm, we have made the following supplements in the Conclusion, and please see lines 500-502 (All Markup): “GBSAED algorithm has a better performance in processing anomalies with spatial characteristics rather than dealing with subpixels. Therefore, our further direction will focus on the adaptability improvement of the algorithm.”
Point 7: References
Additional newer studies (2020 and 2021) could be analyzed. This would extend the Introduction and the Discussion).
Formatting error at line 486, and line 526. Check references and make sure they are in accordance with the formatting guidelines.
Response 7:
Thank you so much for your careful check, we have corrected the formatting errors in the references. At the same time, we checked other parts of the manuscript to ensure that there were no similar errors.
According to your suggestion, we selected two algorithms from the correlation between the research and the newly published articles to conduct experiments, and some other algorithms were introduced in the introduction.
- Cheng, T.; Wang, B. Graph and Total Variation Regularized Low-Rank Representation for Hyperspectral Anomaly Detection. IEEE Transactions on Geoscience and Remote Sensing 2020, 58, 391-406, doi:10.1109/TGRS.2019.2936609.
- Taghipour, A.; Ghassemian, H. Visual attention-driven framework to incorporate spatial-spectral features for hyperspectral anomaly detection. INTERNATIONAL JOURNAL OF REMOTE SENSING 2021, 42, 7454-7488, doi:10.1080/01431161.2021.1959668.
- Li, L.; Li, W.; Du, Q.; Tao, R. Low-Rank and Sparse Decomposition With Mixture of Gaussian for Hyperspectral Anomaly Detection. IEEE TRANSACTIONS ON CYBERNETICS 2021, 51, 4363-4372, doi:10.1109/TCYB.2020.2968750.
- Zhang, Y.; Fan, Y.; Xu, M. A Background-Purification-Based Framework for Anomaly Target Detection in Hyperspectral Imagery. IEEE Geoscience and Remote Sensing Letters 2019, PP, 1-5.
- Ma, Y.; Fan, G.H.; Jin, Q.W.; Huang, J.; Mei, X.G.; Ma, J.Y. Hyperspectral Anomaly Detection via Integration of Feature Extraction and Background Purification. IEEE GEOSCIENCE AND REMOTE SENSING LETTERS 2021, 18, 1436-1440, doi:10.1109/LGRS.2020.2998809.
Author Response File: Author Response.docx
