An Innovative Approach for Effective Removal of Thin Clouds in Optical Images Using Convolutional Matting Model

Round 1

Reviewer 1 Report

This study proposes a new image-matting based method for removing clouds from a single-frame RS image. The proposed SCM-CNN is interesting because it is unique as a cloud removal method and has good validation results. On the other hand, I recommend a major revise of this paper because I think the authors should improve the style of the paper and some descriptions to enhance the value of the paper. I would like to describe some comments which the authors would consider in revising the manuscript.

[A] Overall comments

 (1) The structure of the paper does not follow the format specified by MDPI; please follow MDPI guidelines and clearly separate Materials and Method, Result, and Discussion. For example, section 3.3 is a mixture of Methods, Results, and Discussion. The redundant descriptions for discussion should be organized throughout the paper.

(2) Optical sensor includes not only visible sensors but also infrared sensors. If the application to infrared images is not assumed, the term "visible sensor" should be used instead of "optical sensor" in the title and text.

(3) Although the proposed method is intended for cloud removal, it is only applicable to thin clouds, and many images containing thick clouds are out of scope. Therefore, the phrase "removalen clouds" in the title is not adequate. On the other hand, it could be used to remove haze, though it is not cloud. Based on these, I recommend a more appropriate descriptions in the title and the text.

(4) The difficulty of cloud removal in remote sensing imagery is particularly pronounced in the cryosphere. The proposed method is expected to be less accurate in the cryosphere, which includes Tibet, where the authors point out that there are many clouds. However, there is only a brief mention of snow in the paper, and the Introduction mentions frozen area studies (Line 32) as an application area. Please clearly indicate its applicability and limitations in the cryosphere.

(5) The composition of the dataset used in the training and evaluation has a significant impact on the results of the accuracy evaluation, but there seems to be almost no description of the data composition. For example, what combination of latitude and longitude, surface cover type, season, solar altitude, etc. were used to prepare the background images? Also, how realistic and diverse was the cloud superimposition on these background images? A detailed description of the data composition of the sea images with clouds collected is needed.

(6) The resolution of Figures is generally low. Please improve the quality.

[B] Other comments

L38: Although a type of optical image, thermal image is not similar to human perception.

L43: The statement "making most optical RS products useless" is an extreme statement. Even if the image is covered with cirrus clouds or haze, there are some applications where it can be fully utilized.

L48: "Multi-temporal cloud removal methods" may be able to remove even thick clouds, and in fact have been used in many applications. However, "single-image cloud removal methods" cannot remove thick clouds and can only correct thin cloud areas. Although in general the "multi-temporal" approach is superior in terms of practicality, these two approaches should be considered complementary.

L126-130: The thin cloud effect is actually a combination of a multiplicative factor that attenuates light from below the clouds and an additive factor that adds reflected light at the cloud tops, and equation (1) is a simplified model. This should be clarified.

L140: The threshold value of α for separating thick and thin clouds should be determined based on a comparative assessment of accuracy; what was the basis for the threshold value of 0.5?

L168: There is no explanation for the value of 0.8 in the denominator of Equation 4. This value should also be determined based on a comparative evaluation of accuracy.

L169--: The correspondence between Fig. 2a and the explanation in section 2.3 is confusing. 14 outputs is mentioned in Line 186, but to which part of the figure do these 14 outputs correspond? How do the six arrows in Output correspond to the sub-networks in Backbone? Please write a more detailed explanation in the text and the figure caption.

L219-223: Symbols seem to be removed from the text.

L241: "Algorithm 1" needs a caption.

L265: Please add a description on "S2cloudless masking algorithm" with reference.

L295--: In section 3.3, please describe the hyperparameters given to each method.

Author Response

Dear Reviewer, Thank you for your advice. I have made careful changes and the response is available as a PDF attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Summary of review: The proposed model is based on cloud opacity, a concept that integrates the fact that the ground signal cannot be restored below thick clouds with a single image (ill-posed problem), which is a good thing (some cloud removal articles ignore this…). Experiments and discussions are thorough, yet the manuscript has key limitations. The dataset used in experiments is not clearly described (number of images), and more importantly, it seems to include only RGB bands, which limits the usefulness of the proposed method for most remote sensing applications, which require NIR and or SWIR bands. The proposed model should be extended.

Therefore I recommend a major revision. See attached file for detailed comments.

Comments for author File: Comments.pdf

Author Response

Dear Reviewer, Thank you for your advice. I have made careful changes and the response is available as a PDF attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The manuscript proposes a saliency cloud matting convolutional neural network (SCM-CNN) to address the problem of cloud removal in remote sensing images. Cloud interference is a significant factor that impacts the accuracy of geo-spatial data processing and image analysis. This research topic is therefore of great importance in providing potential support for data analysis. The manuscript presents a comprehensive methodology introduction and experimental analysis. However, some concerns have been raised by the reviewer, which are outlined below:

1. The contributions of the work need to be clarified. While four contributions are listed in Section 1, some of them are not fully discussed. In order to justify the significance of the proposed approach, it is necessary to further prove or explain its contributions to the network or the experiments. For example, the purpose of designing the "channel global max/average pooling" structure should be explained, and its impact on the model's performance should be discussed. The rationale for using a multi-objective loss function be explained.

2. The introduction section should be improved. The reviewer suggests that more preliminary studies related to the cloud detection/removal topic, especially deep learning-based methods, should be introduced.

3. Clearer pictures should be provided for all figures in the draft.

4. In the experimental section, references should be added for the comparative methods.

5. It can be observed that the cloud removal results of SCM-CNN using U2Net are more reliable than using ResNet-50 as the backbone. Further analysis is suggested be provided to explain why U2Net performs better.

6. Some minor errors need to be corrected, such as in line 405, where "... is better than "SCM-CNN(ResNet-50"..." should be revised to "... is better than SCM-CNN(ResNet-50) as the backbone...".

Author Response

Dear Reviewer, Thank you for your advice. I have made careful changes and the response is available as a PDF attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors have responded appropriately to my comments and revised the manuscript, and I believe it is worthy of publication. However, I feel that the following points could still be improved.

I feel that the introduction is too long. Although numerous previous studies are presented in detail, this is not a review article, and I recommend that the authors reduce the description of less relevant literature. Also, I do not think it is very common to give subsections in the introduction. In light of the above, I encourage the authors to rewrite the introduction.

The separation of results and discussion still does not seem sufficient. For example, many of the figures and tables in the discussion part seem to be better placed in the results part. The results should be presented in the results part and the discussion based on the results should be presented in the discussion part.

The following is a minor issue: Put a line break before the chapter title in “2. Materials and Methods”.

Author Response

Dear Reviewer,

Thank you for your feedback. I also agree that the introduction section was too lengthy. With regard to it, I made the following adjustments: 1. The introduction of cloud detection, which is less relevant to this paper, has been removed. 2. Simplified the expression of the methods for cloud removal in time series remote sensing images and single remote sensing images to a great extent. 3. Eliminated the subsections in the introduction.

Furthermore, I made further adjustments to the Results and Discussion section by moving all relevant figures and tables related to the analysis to the discussion part.

Lastly,  I have inserted a line break before the chapter title in "2. Materials and Methods".

I would like to express my sincere gratitude for taking the time to review my paper and providing valuable feedback. I have carefully revised my paper based on your guidance and suggestions.

Throughout this process, your diligent review and comments have made me aware of the shortcomings in my experiments and paper, and have helped me gain more experience and growth. Your guidance and suggestions have greatly contributed to my learning and research, and I will continue to improve my work and incorporate your suggestions into my future work.

I admire your expertise, experience, and critical thinking, all of which are goals I strive for in my research. 

Sincerely,

Renzhe, Wu

Reviewer 2 Report

Thank you to the authors for having addressed my comments and concerns thoroughly, with extensive additional experiments. I can now recommend this manuscript for publication. I still look forward with your continuation work including all spectral bands.

Author Response

Dear Reviewer,

I would like to express my sincere gratitude for taking the time to review my paper and providing valuable feedback. I have carefully revised my paper based on your guidance and suggestions.

Throughout this process, your diligent review and comments have made me aware of the shortcomings in my experiments and paper, and have helped me gain more experience and growth. Your guidance and suggestions have greatly contributed to my learning and research, and I will continue to improve my work and incorporate your suggestions into my future work.

I admire your expertise, experience, and critical thinking, all of which are goals I strive for in my research. Once again, thank you for your review and guidance, and I hope that you will continue to follow my research and provide me with guidance and support in my future studies.

Sincerely,

Renzhe, Wu

Renzhe, Wu

Reviewer 3 Report

This submission has been revised and I have no more comments.

Author Response

Dear Reviewer,

I would like to express my sincere gratitude for taking the time to review my paper and providing valuable feedback. I have carefully revised my paper based on your guidance and suggestions.

Throughout this process, your diligent review and comments have made me aware of the shortcomings in my experiments and paper, and have helped me gain more experience and growth. Your guidance and suggestions have greatly contributed to my learning and research, and I will continue to improve my work and incorporate your suggestions into my future work.

I admire your expertise, experience, and critical thinking, all of which are goals I strive for in my research. Once again, thank you for your review and guidance, and I hope that you will continue to follow my research and provide me with guidance and support in my future studies.

Sincerely,

Renzhe, Wu