Difference Curvature Multidimensional Network for Hyperspectral Image Super-Resolution
Round 1
Reviewer 1 Report
This paper propose a Difference Curvature Multidimensional Network (DCM-Net),and design a difference curvature branch (DCB) and a multidimensional enhanced block (MEB),consisting of several cascaded multidimensional enhanced convolution (MEC) units. In summary, the research is interesting and provides valuable results, but the current document has several weaknesses that must be strengthened in order to obtain a documentary result that is equal to the value of the publication.
Title, Abstract and Keywords:
(1) The abstract is complete and well-structured and explains the contents of the document very well. Nonetheless, the part relating to the results could provide numerical indicators obtained in the research.
Chapter 1: Introduction
(2) On a general level, the study of the proposed detection techniques is reasonable, and the explanation of the objectives of the work may be valid. However, the limitations of your work are not rigorously assumed and justified.
(3) Please check the "3 DCNN" or "3DFCNN" " in line 58.
(4) Please elaborate on the meaning of "HSI SR" in the line 68.
(5) The first paragraph introducing the research topic may present a much broad and comprehensive view of the problems related to your topic with citations to authority references (High-accuracy multi-camera reconstruction enhanced by adaptive point cloud correction algorithm).
(6) Vision technology applications in various engineering fields, should also be introduced for a full glance of the scope of related area. For object detection, please refer to Real-time detection of surface deformation and strain in recycled aggregate concrete-filled steel tubular columns via four-ocular vision; 3D global mapping of large-scale unstructured orchard integrating eye-in-hand stereo vision and SLAM.
Chapter 3: Method
(7) It is recommended to describe the number of train set at the beginning of the article.
(8) The Figure 1 should be placed after “In this section, we present the proposed DCM-Net in details, including the network structure, the Multidimensional Enhanced Block (MEB), the Difference Curvature-based Branch (DCB), and the loss function. The overview network structure of the proposed 153 DCM-Net is illustrated in Fig. 1”.
Chapter 4: Results
(9) Please elaborate on the meaning of "MACs" in table 1.
(10) Please show the loss function data of total loss curve of DCM-Net.
(11) More detailed explainations may be presented in the Figure 4.
Chapter 6: Conclusion
(12) It should mention the scope for further research as well as the implications/application of the study. I recommend including the limitations regarding the consideration of damage indicated in this review in the limitations assessment. This part of the document can be improved and completed with more rigor.
Comments for author File: 
Comments.docx
Author Response
Please see the attachment
Author Response File: Author Response.docx
Reviewer 2 Report
This paper proposes a deep difference curvature-based network with multi-dimensional enhanced convolutions for HSI super-resolution. The topic is interesting and matches well for MDPI Remote Sensing journal. The paper contains review of related works and shows good simulation studies.
However the paper has some unclear points, and the following major and minor concerns.
The authors formulate the main contribution of the paper as follows:
“We devise a multidimensional enhanced convolution (MEC), which leverages a bottleneck projection to reduce the high dimensionality and encourage inter-channel feature fusion, as well as an attention mechanism to exploiting spatial-spectral features. … We propose an auxiliary difference curvature branch (DCB) to guide the network to focus on high-frequency components and improve the SR performance on fine texture details.”
- Thus, the paper proposes a rather complex neural network in structure. It is important to show that this complexity is justified. In section 5, the authors have performed some of the elements of ablation studies. However, the paperwould benefit greatly if this analysis were continued. For example, if the effect on the result of different architectures of the MEС module, Attention-based Guidance and the choice of different options for the loss function would be demonstrated.
- The reasons for choosing the hyperparameters of the neural network architecture are unclear. In particular, it is unclear whether the experiments were carried out with different numbers of neural network layers.
There are also typos in the paper.
Line 59 - And [14? ] also ???
Line 327 - “Our-w/o MC” - “Our-w/o MEC” ???
Line 328 - without MC, - without MEC, ???
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
The authors propose a deep difference curvature-based network with multi
dimensional enhanced convolutions for HSI super-resolution. In particular, they exploit spectral correlation via a self-attention mechanism. The difference curvature operation is useful to extract very detailed information and to remove unwanted noise. Experiments are conducted on three benchmark datasets.
The proposed approach is well described but when it comes to the mathematical formalism I find some confusion. Here are some observations:
- Eq. (1)-(2), what does the symbol Frec stand for? In the text you write that you feed H_DCB and H_SPB with the input low-resolution image ILR, hence I would have expected to see ILR in place of Frec. Otherwise, if there is some preliminary operation, please, specify the details.
- Page 5, ILR(S) : the superscript should be either a small s or, simply, change the symbol for the index. In this way it is clearer that you process each image in the S groups.
- Eq. (3), what is F0(S) ? Again, maybe S-->s, but the main issue here it is the lack of the definition of F0.
- Eq. (11), what does x (small bold x) denote? Is it a band of the input image X (capital bold X) or is it the input image X itself? Please, again, be careful with the symbols you use or introduce.
- Eq. (12), what does the subscript "i" denote? It is not used also in the definition (13) neither in (14). Please, specify the meaning of the index "i".
- Page 7, lines 210,211: please double check the 5 kernels entries. Also, could you please add a more meaningful caption to Fig. (3)? For instance, you could specify what type of kernels do we see by looking at that, i.e., which one is kernel f_x, f_y, f_xx, etc. etc.
- Eq. (16), add the index "1" as a subscript to the considered norm.
- Fig. (6) and Fig. (8) I would add a colorbar indicator.
Author Response
Please see the attachment
Author Response File: Author Response.docx
Reviewer 4 Report
This study focused on the proposing of Difference Curvature Multidimensional Network for
Hyperspectral Image Super-resolution. While this study is generally adequately structured and makes sense, a major upgrade of the remote sensing component is necessary (as this study dominantly focuses on the computer science part of the research and some crucial remote sensing components regarding data acquisition and discussion is almost completely disregarded). The Discussion section needs to be thoroughly restructured and improved, as well as the majority of figures in the manuscript.
Section 2 is misplaced as an individual section and should be shortened and integrated into Introduction. The authors must consider that the text in all figures should be clearly visible to the readers. That means the font size should be increased in all figures almost to match font size of the surrounding text. I suggest that you increase the size of some figures as well. Please rename section 3 to Materials and Methods and move current subsections 4.1 and 4.2 to previous section. These subsections should not be placed under a Result section.
More thorough explanation of input remote sensing data is necessary, alongside the information regarding imaging sensor, carrier, spatial resolution, imaging relative altitude etc. After all, this is the Remote Sensing journal and this information is highly significant to its readers.
The actual Discussion is almost non-existent in terms of comparison of these results and results from previous studies. You used zero (?!) references in the Discussion section. Moreover, Figure 11 and Table 6 should belong to the Results section. A complete upgrade and expansion of the Discussion section is required before publishing in my opinion.
Specific comments:
Abstract: I suggest the authors to replace “visible light images” with “RGB images” where it is applicable (in the rest of the paper as well).
Line 38: Formatting error. Please fix it.
Figure 4, 11: You should insert a scalebar in these figures. Writing a scale in text is not preferrable.
Figure 8: Please use other colors as currently these subfigures are very poorly visible.
Author Response
Please see the attachment
Author Response File: Author Response.docx
Round 2
Reviewer 4 Report
The authors addressed all my comments in this version of the manuscript. I have no more suggestions to add.
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
