Rapid Detection of Iron Ore and Mining Areas Based on MSSA-BNVTELM, Visible—Infrared Spectroscopy, and Remote Sensing
Round 1
Reviewer 1 Report
In this paper, the authors present a rapid detection method for total iron content based on Visible-Infrared Spectroscopy, MSSA-BNVTELM and Remote Sensing. This detection method is categorized into ore detection and mine detection based on the type of model input. Good attempt has been made by the authors in proposing the use of machine learning method for reliable ore and mine detection. And the experiments also proves the feasibility of the method in this paper. In my opinion, this paper is well written and worth publishing. However, before reconsidering whether it is suitable for publication, some minor modifications need to be made. These problems are as follows:
（1）In wavelet transform, the author claims that the curve after wavelet transform is smoother, but I did not see the spectral curve before wavelet transform. Therefore, the author should combine spectral curves for comparison, rather than giving conclusions.
（2）The y-axis in Figure 6 should be "Cumulative contribution rate" rather than "Contribution rate".
（3）The formula for the Pearson coefficient (R2) is not defined in this paper.
（4）In the experiments of this paper, the authors evaluated the models based on RMSE and R2. The RMSE metrics can only evaluate the gap between the models and not the accuracy of the models, because different data outlines can cause a large gap in the RMSE metrics. Therefore, I suggest that the authors add a graph comparing the experimental results with the actual values.
（5）The formulas are not italicized, please carefully check the article.
No Comments.
Author Response
Please refer to the attached document
Author Response File: 
Author Response.docx
Reviewer 2 Report
The authors have done an interesting job on the rapid identification of Iron ore in a mining area. The application of ML algorithms in processing spectral data for detecting Iron ore is novel and can be an interesting topic for researchers. However, I decided to reject this manuscript because of its serious problems. It actually needs a comprehensive overhaul which totally will change the current structure of the manuscript. Below you can find my comments:
The English language of the manuscript is very weak and should definitely be improved before further submissions.
Introduction:
The authors should point to the spectral difference between Iron ore and other types of Iron like gangue which is commercially worthless. How spectroscopy can distinguish between the two types?
How remote sensing can be used for detection of the Iron ore? What is the main spectral feature of the Iron ore which makes point and imaging spectroscopy able to recognize it?
Materials and methods:
You should introduce your study area by providing a map so that the readers can have an idea about the scale and coordinates of your study.
The section is mostly a very long and boring theory about the methods used in this study. It should certainly be revised to a more concise format.
There is no methodology for the remote sensing part of the manuscript. The data used, pre-processing, and processing methods are missing in the manuscript.
Results and discussion:
This section is only reporting the results. There is no discussion on the reason why some detection methods perform better than others. A comprehensive discussion should be added considering the results obtained by other studies.
How and why is the remote sensing inversion done? What was the inversion method?
Considering the 30m pixel size of the Landsat images, how do the authors justify the representativeness of a single point measurement for a 900 m2 area?
Conclusion:
This part should only report the main results which have fulfilled the objectives of your study.
The English language of the manuscript is very weak and should definitely be improved.
Author Response
Please refer to the attached document
Author Response File: Author Response.docx
Reviewer 3 Report
This paper evaluated the Spectroscopy and Landsat-8 multispectral data coupled with an Extreme learning machine (ELM) to build detection models of Iron Ore in the mining Area. The topic was interesting, but the paper needs more work and clarification.
1. Line 14 : « Therefore, this paper proposes a method of TFE detection based on reflection spectroscopy and remote sensing. ». Please, specify the spectral range.
2. Line 16 : « obtained by spectral experiment ». Please clarify.
3. Line 22 : « an inversion model ». Please explain.
4. Lines 38-39 : « Spectroscopy detection can realize fast detection and non-destructive detection and has been widely used in food detection [5], agricultural detection [6], soil detection [7], ore detection [8], and medical detection [9]. », and Lines 53 : « Remote sensing has been successfully applied in soil detection [15], water quality detection [16], slope detection [17], mine detection [18], and marine detection [19]. ». Please delete. Focus on the research works using spectroscopy and Remote Sensing for the Detection of Iron Ore and Mining Areas.
5. In the last paragraph of the introduction, please explain the work's originality.
6. Add a figure of the location of the study area and the 200 iron ore samples.
7. Add the dates of samples acquisition and Landsat data.
8. Line 93 : « After performing five spectral experiments, the TFE of the powder was determined using potassium dichromate titration. ». Please add the bibliographic reference of the method. Explain how to perform the assay of the total iron content (TFE).
9. Line 92 : « The band range was 340 nm- 2500 nm. ». It is the visible to near infrared (VNIR) and short-wave infrared (SWIR) spectroscopy; AND not the infrared Spectroscop, as stated in the title of the paper.
10. How was the geographic coordinate obtained for each sample?
11. Describe the Landsat data processing steps.
12. Why the choice of the wavelet transform and principal component analysis methods for data smoothing and dimensionality reduction?
13. Please add the bibliographic reference of the VTELM and SSA algorithms. Explain the choice of these algorithms.
14. Please add the three evaluation indexes, namely : determination coefficient (R2), and Mean Absolute Error (MAE), and the ratio of performance to interquartile (RPIQ). https://doi.org/10.1016/j.trac.2010.05.006 and https://doi.org/10.1007/s11356-022-21890-8
15. The cross-validation technique or independent validation samples was used for assessing the models' performance ?
16. The uncertainty of the predicted models must be quantified in order to confirm these results. However, I suggest adding a Models uncertainty analysis. Pls, See the section Models uncertainty analysis: https://doi.org/10.3390/rs14164080
17. Based on Spectroscopy results what are the most important layers for TFE detection?
18. what is the influence of the low number of spectral bands of the Landsat-8 data and the spatial resolution of 30m on the accuracy of the TFE results of the mine area?
19. In the discussion section, compare and interpret the results with other work using other types of remote sensing data or other algorithms.
Regards, reviewer
Author Response
Please refer to the attached document
Author Response File: Author Response.docx
Reviewer 4 Report
The manuscript reviewed (id remotesensing- 2518873) is entitled " Rapid Detection of Iron Ore and Mining Areas Based on MSSA-BNVTELM, Visible-Infrared Spectroscopy, and Remote Sensing".
The authors investigated the total iron content“ based on reflection spectroscopy and remote sensing“ using a MSSA-BNVTELM (the authors do not explicitly explain what the acronym stands for) based on a “two hidden layer extreme learning machine with variable neuron 18 nodes based on an improved sparrow search algorithm and batch normalization optimization”. The method is applied on Landsat data.
The bibliography on the methodological aspects in the Introduction is adequate.
The methodological aspect is well presented. Although the approach is generally not innovative, it is interesting to see its application to this specific topic.
However, there are some serious issues to be addressed:
1) The authors do not provide any information on the study area
2) What do the authors mean with total iron content? How is this quantified? Do they mean ferrous and ferric content?
3) The remote sensing part is not adequately described, in particular concerning the processing of the data. The authors address this only in lines 99-101 and 287-293. Especially in lines 287-293, the authors mention “The remote sensing data are processed and analyzed by atmospheric correction and other methods to obtain the remote sensing image of the mine area”. What are the other methods? What type of atmospheric correction was used? Did they encounter any striping issues?
4) In Figure 9b, there is no explanation on what the y-axis represents for (units?).
5) I do not see how the authors came to any of the observations/conclusions they refer to in lines 294-301.
6) I do not understand the reason why the authors selected to use Landsat 8 OLI data. It is well known that iron diagnostic features are observed within the Visible-Near Infrared (VNIR) region of the E/M spectrum. The more spectral information is available within this spectral area, the better the results will be. It would be preferable for the authors to apply their method on more suitable data such as for example Sentinel-2 (they are free to download and most of the datasets are already provided in Level 2, meaning already atmospherically corrected). The advantages are that Sentinel-2 acquires information at 12 spectral bands (after atmospheric correction) within VNIR-SWIR (10 bands in VNIR and 2 in SWIR), adds new significant capabilities in terms of revisit time, of spatial resolution (10m) and of spectral information compared to the Landsat series. Especially for the latter, Sentinel-2 provides 10 spectral bands in the VNIR region (suitably centered to detect both ferrous and ferric presence) compared to Landsat 8 OLI’s five VNIR bands. This is very important in prospecting iron ores or minerals and generally in detecting iron presence. I strongly recommend to the authors to apply their methodology to this type of data and evaluate the results.
7) I would suggest (it is not mandatory) to the authors to additionally compare their results with some more conventional/simple approaches for mapping iron presence such as Spectral Indices. The authors can find relative Information in Van der Meer et al., 2014 and Ge et al., 2020 as well as in the Index Database for remote sensing indices (https://www.indexdatabase.de/db/r.php)
Considering the above, I cannot recommend the publication of this manuscript as it is in its present form. I strongly recommend to the authors to take into consideration my aforementioned comments and re-submit their manuscript.
The English language requires moderate revision.
Author Response
Please refer to the attached document
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
Dear authors,
Thanks for your efforts in addressing most of my concerns regarding this manuscript. Below please find some minor considerations:
1- "Reflection spectroscopy" should be "reflectance spectroscopy" throughout the whole manuscript.
2- Revise this sentence in the abstract. It is not fluent: "Furthermore, this paper used Sentinel-2 data with MSSA-BNVTELM to construct a detection model of the mine area to detect the TFE of the whole mine area."
3- Revised the last sentence of the abstract to "The results show that the remote sensing of the mine area can be useful for detection of the TFE distribution, providing assistance for the mining plan."
4- Revised lines 431 to 433 of the conclusion to: "In addition, this paper utilizes Sentinel-2 and Landsat-8 data to establish TFE detection models in the mining area."
English language needs minor modifications
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
Good job
Author Response
Thank you for recognizing the manuscript and for the work done on it!
