Fast and Accurate Hyperspectral Image Classification with Window Shape Adaptive Singular Spectrum Analysis

Round 1

Reviewer 1 Report

I think you should claery seperate the method, results and discussion. This will enhance readability of your work. Also try to improve on the conclusion so that readers can easily identify the essence of the work. 

The English language usage is appropraite. Though can also be improved.

Author Response

Reviewer: 1

Comments to the Author

I think you should clearly separate the method, results, and discussion. This will enhance the readability of your work. Also, try to improve on the conclusion so that readers can easily identify the essence of the work.

Response: Modifications have been incorporated in the original text.

The English language usage is appropriate. Though can also be improved.

Response: The use of the English language has been revised.

Reviewer 2 Report

1.There is a need to unify the past or present tense in Introduction, add some methods of analysing feature extraction. SPSTT10.1109/TGRS.2023.3307071,CGCRD10.1109/TGRS.2022.3169171.

2.The contribution of the paper needs to be recapitulated and summarised better in 3 articles.

3.Figures 6,7 of the paper are not very clear, it is suggested to change them to 600 dpi and add some descriptions to Figures 1,2.

4.Can the spatial spectral features of the papers be visualised for comparison.

5.Increasing the time complexity analysis of the algorithm

6.The paper's formulae are not neatly numbered and it is recommended that they be rearranged. And some of the formulas in the paper are in red.

Minor editing of English language required.

Author Response

Reviewer2

1.There is a need to unify the past or present tense in Introduction, add some methods of analysing feature extraction. SPSTT10.1109/TGRS.2023.3307071,CGCRD10.1109/TGRS.2022.3169171.

Response：The tense issue has been addressed in the document. The methods for analyzing feature extraction have already been added in the original text.

2.The contribution of the paper needs to be recapitulated and summarised better in 3 articles.

Response：The contribution of the paper has been summarized in 3 points.

3.Figures 6,7 of the paper are not very clear, it is suggested to change them to 600 dpi and add some descriptions to Figures 1,2.

Response : Image has been updated to a clearer version. The detailed description of Figure 1 is provided in the main content of Chapter 3, while the detailed description of Figure 2 is presented in the second section of Chapter 3.

4.Can the spatial spectral features of the papers be visualised for comparison.

Response：Utilizing three forms of visualizing feature mappings, no significant differences are discernible in the feature mapping graphs. It is more persuasive to compare subjective classification results with objective classification data.

5.Increasing the time complexity analysis of the algorithm

Response: In the experimental analysis section of the paper, it is described that the current spectral analysis methods suffer from high computational complexity and long runtime. This is attributed to the use of window traversal for each pixel, leading to significant redundant computations. Additionally, to achieve high-precision classification, integration of various existing algorithms further amplifies the computational load. The proposed method in this paper eliminates the window traversal operation, reducing redundancy and computation time. Moreover, the improved feature extraction method enhances classification accuracy without the need for combining with other algorithms.

6.The paper's formulae are not neatly numbered and it is recommended that they be rearranged. And some of the formulas in the paper are in red.

Response: The formula has been modified in the original text.

Minor editing of English language required.

Response: Modifications have been made in the original text.

=========================================================================

Author Response File: Author Response.docx

Reviewer 3 Report

Comments for author File: Comments.pdf

The English quality is  poor. Please make proper use of grammar and punctuation. 

Author Response

Reviewer3

This study addresses a significant and relevant problem in remote sensing image processing—hyperspectral image classification. It introduces a novel spectral-spatial feature extraction method, WSA-SSA, to reduce computational complexity. It highlights the classification accuracy (97.56%, 98.34%, and 99.77%), and the significant improvement in classification speed compared to other methods adds credibility to the research. The research's practical implications are evident, as it improves classification accuracy and significantly reduces computational time, making it applicable in real-world scenarios. However, I have some concerns regarding the manuscript's English writing; the basic sections, like Material and methods, Results, and Discussion, are not well organized. It seems messy regarding manuscript arrangement and basic structure. The dataset is not enough and English language must be revised professionally.

Response: Thank you for your general comments.

Line 14-16: The introduction, problem and study background is not appropriate, repeat it please. Make use of connecting and supporting words between sentences in the abstract and whole manuscript.

Response: Modifications have been made in the original text.

Line 26: Which three datasets?

Response: Modifications have been made in the original text.

Line 31: Introduce acronyms in each section

Response: The acronyms were explained upon their first mention.

Line 31: “THE nearly continuous spectral” This is not appropriate It might be “The continuous spectral bands” Revise in the whole manuscript

Response: Modifications have been made in the original text.

Line 34,35: Repeat sentence, it is not appropriate

Response: Modifications have been made in the original text.

Line 45,46: Revise sentence and recheck grammar, please.

Response: Modifications have been made in the original text

Line 64: Use same abbreviation, uniformly HSI or HSIs

Response: Throughout the entire document, the term has been standardized to 'HSI'

Line 99-116: Enlist the objectives correctly, and shortly in precise manner. Figure 3,4 and 5: Unable to read legends

Response: Modifications have been made in the original text.

Line 255: Mention size dimension in table

Response: The "size" column in Table 1 specifies the size dimensions for the three datasets used.

Figure 6,7: Blurry, difficult to read

Response: Figures 6,7 have been updated to a clearer version.

=========================================================================

Author Response File: Author Response.docx

Reviewer 4 Report

The manuscript presented a hyperspectral image classification method, which utilized proposed WSA-SSA method to extract spatial-spectral feature from superpixels, and performed final classification with SVM. The major issues of the manuscript are the novelty and the final performance.

1. The hyperspectral image classification has been well studied, particularly using the Indian Pine, Salinas, and U of Pavia datasets. To evaluate the performance of your classification approach, it is essential to compare it with other state-of-the-art methods. Such as EPF-based method as following, this method utilized 10%, 2%, and 6% of the pixels as training samples from the Indian Pine, Salinas, and the U of Pavia, in contrast to the 8% used in your referenced literature.

X. Kang, S. Li and J. A. Benediktsson, "Spectral–Spatial Hyperspectral Image Classification With Edge-Preserving Filtering," in IEEE Transactions on Geoscience and Remote Sensing, vol. 52, no. 5, pp. 2666-2677, May 2014.

The following paper is also based on ERS superpixel method and achieved state-of-art performance.

L. Fang, S. Li, W. Duan, J. Ren and J. A. Benediktsson, "Classification of Hyperspectral Images by Exploiting Spectral–Spatial Information of Superpixel via Multiple Kernels," in IEEE Transactions on Geoscience and Remote Sensing, vol. 53, no. 12, pp. 6663-6674, Dec. 2015.

2. The evaluation metrics OA and AA in your manuscript access your classification performance from only a single perspective, other measures that evaluate performance from misclassification point of view, such as precision or false alarm rate should also be considered.

B. Xue et al., "A Subpixel Target Detection Approach to Hyperspectral Image Classification," in IEEE Transactions on Geoscience and Remote Sensing, vol. 55, no. 9, pp. 5093-5114, Sept. 2017.

J. Li, B. Xi, Y. Li, Q. Du, and K. Wang, “Hyperspectral classification based on texture feature enhancement and deep belief networks,” Remote Sens., vol. 10, no. 3, p. 396, 2018.

3. In section 3.2, which is the most important part to demonstrate the novelty of your work, contains a description of your proposed WSA-SSA method from lines 196 to 206. However, this description, including the notation system used, is unclear and should be revised for better clarity.

4. What procedures were included in your documented running time in table 2 to 4? It seems that the training time of the SVM is not included in, please clarify.

5. In tables 2 to 4, consider replacing the columns indicating the percentage of training and testing set samples with columns indicating the actual number of training and testing samples.

6. Please review the manuscript's formatting, such as the title of table 4 and references.

Author Response

Reviewer4

The manuscript presented a hyperspectral image classification method, which utilized proposed WSA-SSA method to extract spatial-spectral feature from superpixels, and performed final classification with SVM. The major issues of the manuscript are the novelty and the final performance.

Response: Thank you for your general comments.

1. The hyperspectral image classification has been well studied, particularly using the Indian Pine, Salinas, and U of Pavia datasets. To evaluate the performance of your classification approach, it is essential to compare it with other state-of-the-art methods. Such as EPF-based method as following, this method utilized 10%, 2%, and 6% of the pixels as training samples from the Indian Pine, Salinas, and the U of Pavia, in contrast to the 8% used in your referenced literature.

Response: The comparison methods selected in this paper are all hyperspectral classification methods based on singular spectrum analysis. Through comparison, it can be observed that the improvements made to the singular spectrum analysis algorithm in this paper not only result in high-precision classification but also significantly enhance the classification speed.

2014. Kang, S. Li and J. A. Benediktsson, "Spectral–Spatial Hyperspectral Image Classification With Edge-Preserving Filtering," in IEEE Transactions on Geoscience and Remote Sensing, vol. 52, no. 5, pp. 2666-2677, May 2014.

The following paper is also based on ERS superpixel method and achieved state-of-art performance.

2015. Fang, S. Li, W. Duan, J. Ren and J. A. Benediktsson, "Classification of Hyperspectral Images by Exploiting Spectral–Spatial Information of Superpixel via Multiple Kernels," in IEEE Transactions on Geoscience and Remote Sensing, vol. 53, no. 12, pp. 6663-6674, Dec. 2015.
2016. The evaluation metrics OA and AA in your manuscript access your classification performance from only a single perspective, other measures that evaluate performance from misclassification point of view, such as precision or false alarm rate should also be considered.

Response: In deep learning and machine learning, it is often necessary to assess the quality of classification through metrics.

We make the following definitions: "true" and "false" indicate whether there is a match between the true and predicted values, where "true" signifies that the predicted value aligns with the true value, and "false" indicates a discrepancy between the predicted and true values.

Positive samples predicted as positive (TP),

Negative samples predicted as positive (FP),

Positive samples predicted as negative (FN),

Negative samples predicted as negative (TN).

The evaluation metrics for hyperspectral classification are derived from the confusion matrix, which is calculated based on TN, FN, TP, and FP. Therefore, OA (Overall Accuracy), AA (Average Accuracy), and the Kappa coefficient can objectively evaluate the classification performance.

2017. Xue et al., "A Subpixel Target Detection Approach to Hyperspectral Image Classification," in IEEE Transactions on Geoscience and Remote Sensing, vol. 55, no. 9, pp. 5093-5114, Sept. 2017.
2018. Li, B. Xi, Y. Li, Q. Du, and K. Wang, “Hyperspectral classification based on texture feature enhancement and deep belief networks,” Remote Sens., vol. 10, no. 3, p. 396, 2018.

3. In section 3.2, which is the most important part to demonstrate the novelty of your work, contains a description of your proposed WSA-SSA method from lines 196 to 206. However, this description, including the notation system used, is unclear and should be revised for better clarity.

Response: Modifications have been incorporated in the original text.

4. What procedures were included in your documented running time in table 2 to 4? It seems that the training time of the SVM is not included in, please clarify.

Response: The recorded running time in this paper represents the time required from importing the data to completing the classification, including both the training time and detection time of SVM.

5. In tables 2 to 4, consider replacing the columns indicating the percentage of training and testing set samples with columns indicating the actual number of training and testing samples.

Response:Using percentages to represent training and testing samples is more intuitive.

6. Please review the manuscript's formatting, such as the title of table 4 and references.

Response: Modifications have been incorporated in the original text.

Author Response File: Author Response.docx

Reviewer 5 Report

The paper analysed the inherent features of HSIs. It proposed a novel spectral-spatial feature extraction method called window shape adaptive singular spectrum analysis (WSA-SSA) to reduce the computational complexity of the feature extraction. Combining similar pixels in the neighbourhood to reconstruct every pixel in the window, the main steps are as follows: rearranging the spectral vectors in the irregularly shaped region, constructing an extended trajectory matrix, and extracting the local spatial and spectral information of HSI while removing the noise.

1. The Quality of figures should be improved.

2. Correct the title of Table 4.

3. The article proposed improving existing methods. It is a small piece of knowledge that adds to global wisdom. The references could be improved with a similar one. In conclusion, the authors should emphasize their contribution.

The work is on a good scientific level. I recommend publication.

Only formulas should be correct. 

Author Response

Reviewer5

The paper analysed the inherent features of HSIs. It proposed a novel spectral-spatial feature extraction method called window shape adaptive singular spectrum analysis (WSA-SSA) to reduce the computational complexity of the feature extraction. Combining similar pixels in the neighbourhood to reconstruct every pixel in the window, the main steps are as follows: rearranging the spectral vectors in the irregularly shaped region, constructing an extended trajectory matrix, and extracting the local spatial and spectral information of HSI while removing the noise.

Response: Thank you for your general comments.

1. The Quality of figures should be improved.

Response: images have been updated to a clearer version.

2. Correct the title of Table 4.

Response: Modifications have been incorporated in the original text.

3. The article proposed improving existing methods. It is a small piece of knowledge that adds to global wisdom. The references could be improved with a similar one. In conclusion, the authors should emphasize their contribution.

Response: The conclusion has been modified in the original text.

The work is on a good scientific level. I recommend publication.

Only formulas should be correct.

Response: Modifications have been incorporated in the original text.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

In line 49 "curse"? check it again

In line 77-79...staement needs to be improved. 

In the conclusion, at least, you should mention one/two of the most important findings, not a general statement as you have done in lines 427-430, in which you mention the proposed algorigthm, and others. In the following statement (Line 431) which results are you talking about? Be specific. Then, also in that line 431, your grammar is wrong! It should be "These results are...." 

In fact, the following sentences after that are also not grammatically sound.

You can see for yourself that your paper actually needs a THOROUGH English Language editing!

Work on it before the paper can be published.

The author made efforts to address my concerns, however, the paper still needs thorough English editing before it can be publisged. 

Author Response

Comments and Suggestions for Authors In line 49 "curse"? check it again Response: "Curse" in this context, it more refers to "issue," "challenge," or "dilemma." In the field of computer science, "curse of dimensionality" signifies the problems or challenges arising from an increase in dimensions. Therefore, in this context, "curse" does not literally mean a curse. In lines 77-79... statement needs to be improved. Response: Modifications have been incorporated in the original text. In the conclusion, at least, you should mention one/two of the most important findings, not a general statement as you have done in lines 427-430, in which you mention the proposed algorithm, and others. Response: Modifications have been incorporated in the original text. In the following statement (Line 431) which results are you talking about? Be specific. Then, also in that line 431, your grammar is wrong! It should be "These results are...." Response: This phrase "result in" indicates a consequence or outcome of a certain action or event. In fact, the following sentences after that are also not grammatically sound. You can see for yourself that your paper actually needs a THOROUGH English Language editing! Work on it before the paper can be published. Response: Modifications have been incorporated in the original text. Comments on the Quality of English Language The author made efforts to address my concerns, however, the paper still needs thorough English editing before it can be published. Response: Modifications have been incorporated in the original text.

Author Response File: Author Response.pdf

Reviewer 3 Report

The authors have properly revised the manuscript.

Author Response

Comments and Suggestions for Authors

The authors have properly revised the manuscript.

Response: Thank you very much for your checking work.