Sub-Pixel Mapping Model Based on Total Variation Regularization and Learned Spatial Dictionary

Round 1

Reviewer 1 Report

In this study, authors propose a subpixel mapping model based on total variation regularization and learned spatial dictionary hyperspectral remote sensing data. Paper is well-written, approach is well-described and results, based on three experiments, are convincing so that it is almost ready to be published. I recommend the following adjustments:

1. The objective of the study needs to be stated more straightforward and more consistently in the Abstract and in the end of the Introduction section. Something like: In this study, we propose a novel sub-pixel mapping process based on J-SVD dictionary learning algorithm and total variation as a spatial regularization parameter.
2. The numerical sequence of figures and tables must be checked throughout the text. For example, Figure 1 is cited in L116; Figure 5 is cited in L128; and Figure 4 in L134. Please, cite them in numerical order Same for tables.
3. Figures 6, 7, and 8 can be joined as Figure 6 since they refer to the same location.
4. In the References section, the Remote Sensing journal uses abbreviated form for journal names (e.g., Remote Sens. Environ. instead of Remote Sensing of Environment).

Author Response

we want first to thank you for your precious and valuable comments which enabled us to improve the manuscript structurally and scientifically. We made the necessary corrections as recommended. 

"Please see the attachment."

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper proposed a sub-pixel mapping technique relying on K-SVD dictionary learning (SMKSVD-TV), which adopts total variation as a spatial regularization parameter, to deal with some known problems in remote sensing by having more spatial configurations at the sub-pixel level.

1. The cases of the title were not consistent, where “regularization” should be capitalized as “Regularization”, and “learned” should be “Learned”.
2. The Abstract is a little too verbose. I suggest the authors to improve it and clarify the problem, where the method and the results can support your contributions. There is need to present too many background reviews in the Abstract.
3. The review of the current work are not yet adequate. I suggest you study the sota dictionary learning such as the following (but not limited to) work:
   [1] Shaoning Zeng, Bob Zhang, Jianping Gou, Yong Xu (2021). Regularization on Augmented Data to Diversify Sparse Representation for Robust Image Classification. IEEE Transactions on Cybernetics
   [2] Jianhang Zhou, Shaoning Zeng, Bob Zhang (2020). Two-stage Knowledge Transfer Framework for Image Classification. Pattern Recognition.
4. It would be better if the notations in Subsection 2.1 are shown in a table.
5. In Table 1, all numbers should 2 decimal places, e.g., 58.3 to 58.30.
6. What is the meaning of using bold numbers in the tables?
7. Please pay attention to enhance the writing or use some English writing service to improve the paper. For example, in the Abstract, “To solve inverse problem, …” should be “To solve the inverse problem, …”.

Author Response

"Please see the attachment."

Author Response File: Author Response.pdf

Reviewer 3 Report

This article proposes a good method(SMKSVD-TV) for SPM correction, and the research results also proved that it has better performance than the other three algorithms, especially in localized graphics presentation. Here are some suggestions for amendments to this article:

 

1. It is necessary to systematically present the definition of each code, because this article used a large number of mathematical formulas and codes to explain the calculation process. It is recommended to organize all codes with a chart and explain them. Besides, the nouns used in the research are suggested to be unified (such as Kappa coefficient and Kappa indices).
2. It is recommended that each figure must have descriptions and mark the corresponding position.
3. The explanation about sparse modeling of high spatial resolution data using the dictionary must be more detailed, especially the process of converting Figure 3 to Figure 4, and the definition of dictionary atoms.
4. The accuracy of this study has reached a good level in different regions, but the Kappa coefficient does not seem to reach a better level in some regions. This is worth explaining.
5. The research part uses many manually entered parameters, and it seems cumbersome to set up operations before operation. Is it possible to import a CNN-like operation in this part (repetitive training with training data to gradually reach the best solution and converge work results?)
6. The classification results are currently presented in tables, but what we are more interested in is the location and spatial distribution characteristics of the commission error and omission error. If the location where these errors occur can be marked, it may be possible to use other similar landscape indicators to explain their occurrence.

Author Response

"Please see the attachment."

Author Response File: Author Response.pdf