Convolutional Neural Network Chemometrics for Rock Identification Based on Laser-Induced Breakdown Spectroscopy Data in Tianwen-1 Pre-Flight Experiments

Round 1

Reviewer 1 Report

This work develops a convolutional network to improve rock classification on the LIBS spectra acquired by MarSCoDe for China’s Tianwen-1 Mars exploration mission. The proposed method is compared with LR, SVM and LDA and obtains better results. The application is of course very important. However, the method is lack of innovation in methodology and has no special processing of the data used. Below, please find a summary of my major and minor concerns.

1)    Many works have employed convolutional networks on different LIBS datasets. What is the methodological innovation of this work? Are the pre-processing and algorithm specially designed for the data used?

2)    It is not mentioned how the network structure is chosen. Is it chosen with validation or empirically decided?

3)    I suggest that the sentence “The partitioning of Dataset I is in random mode” should be better explained. How the authors ensure that the samples of the 12 categories are balanced distributed in the training set and validation set. It is possible that the validation set misses a category if the total dataset is randomly divided.

4)    Figure 5 (b) cannot show the main effect or difference of convolutional layers and max-pooling layers and has no benefit to illustrate the principle of the network layers. I suggest that the figure should be better designed or deleted.

5)   The convolutional layers are 1-dimensional. I suggest that ‘kernel sizes are set to 51’ in Line 193 should be corrected to ‘kernel length is set to 5’.

6)    The units of y axis in Figure 6 and Figure 10 should be given.

Author Response

Dear Reviewer,
Thank you very much for your great review work. According to your reasonable and insightful comments, we have made corresponding modifications. The attached PDF file is the complete reply letter. We hope you are satisfied with our replies.

Author Response File: Author Response.pdf

Reviewer 2 Report

The manuscript titled “Convolutional Neural Network Based Laser-induced Breakdown Spectroscopy Chemometrics for Rock Identification in Mars Exploration” by Fan et al.,  is well suited for this journal. Below are the major comments to be furnished by the authors.

·       The title needs to be modified.

·       In figure 3, laser is not shown, figure should be modified.

·       How the compositions were measured, please include the source, and how they determine?

·       Section 2.3 and 2.4 should be after 2.5 LIBS measurements

·       Line 234, what does it mean by “the identical position of each rock sample,”, how much spectral average is considered?

·       The repetition rate which is utilized for the measurements is not provided.

·       How many number of pulses are used at a single spot on the sample? Why they do not consider cleaning shots?  

·       It would be better to include the basic difference between the selected four chemometric models, what information from the spectra they utilize, convert and etc.,. In absence of this, the manuscript looks like just a black box data processing publication. The paper lacks spectroscopic discussion, that is essential for remote sensing journals.

·       Figure 8 is not clear what scores are related to (However the symbols are used)? Please elaborate in the label, the corresponding labels! Maintain in all figures and tables.

·       Figure 9 should be after the experimental diagram.

Please include all representative LIBS spectra in subplot form, marking the elemental and molecular peaks (in the supplementary information)

After evaluating the CNN,

1. why Kernel size is set as 5  X 1? is there any hyper parameter tuning is done to choose 5 X 1? (line 192-193)

2. why 8 Kernels in layer1 and 32 kernels in layer 2 ? is there any significance? (Line195-196)

3. performance measures precision, recall and F1 score is presented only for validation and testing but not for training, please provide the comprehensive data in supplementary details.

4. the impact of the  kernel size and number of kernels are not discussed in the analysis

 

Though I am not a native English speaker if any phrase sounds inappropriate, I have just marked it. The following are my observations:

1.     Line 20, “with some of which”

2.     Line 26,“ the CNN can achieve”

3.     Line 62, “LIBS technique is exactly good”

4.     Line 135, “  Equipped on Zhurong rover”

5.     Line 238, “checking the training effect”

6.     Line 215-216,” implemented by Scikit learn”,

7.     Line 283, “without LIBS laser activation”

8.     Line 307,” the maturity of each model”

9.     Line 308, ” to formally test their”

10.   Reduce the paragraph 317-320, too long sentence

 

11.  Line 327, “The significant imprecision in these predictions are..”

Author Response

Dear Reviewer,
Thank you very much for your great review work. According to your reasonable and insightful comments, we have made corresponding modifications. The attached PDF file is the complete reply letter. We hope you are satisfied with our replies.

Author Response File: Author Response.pdf

Reviewer 3 Report

Great paper, only a few small comments:

1. Suggest replacing “them” with “precision and recall”

2. This phrasing is odd, suggest something simpler: “…and the CNN was still the best performing model.”

3. Suggest inverting this color ram so smaller numbers are on the left

4. Suggest using km not m, or marking the thousands with a comma

5. Excellent point!

6. Phrasing is odd again, perhaps “one of the most popular”

7. Your tense shifted from past to present. Suggest “were”, and checking the rest of this section for consistency.

8. Another excellent point!

Comments for author File: Comments.pdf

Author Response

Dear Reviewer,
Thank you very much for your great review work. According to your reasonable and insightful comments, we have made corresponding modifications. The attached PDF file is the complete reply letter. We hope you are satisfied with our replies.

Author Response File: Author Response.pdf

Reviewer 4 Report

Dear Authors,

Overall good work. However, the validation (60) set size shall be increased or /and compared for the varied results.

Sentences with "we"/"our" shall be rephrased without "we"/"our", which are implicit in your manuscript. Sentences with "In"/"For" in the beginning of sentences may also be improved.

"Fashionable" (line 186) may be suitably replaced with more common words (common/usually).

Conclusion may be order properly in 1-2 paragraphs with clear conclusion.

 

best wishes,

Author Response

Dear Reviewer,
Thank you very much for your great review work. According to your reasonable and insightful comments, we have made corresponding modifications. The attached PDF file is the complete reply letter. We hope you are satisfied with our replies.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

All my questions have been answered.