A Novel Method for Evaluation of Ore Minerals Based on Optical Microscopy and Image Analysis: Preliminary Results
, Andrea Aquino 2,3, Giulia Domenighini 1 and Stefano Pagnotta 2,*
Round 1
Reviewer 1 Report
Overall comments :
The authors describe very well from abstract to conclusion the problematic of time consuming technics applied in mineralogy (especially with SEM-EDX...).
However, the authors are not referencing the LIBS technology applied for mineralogy (LIBS+AI=Automated mineralogy). This new technic is ultrafast for multielemental hyperspectral imaging of rocks, core and chips (including light elements and especially for critical minerals).
The authors should review the literature and report at least some papers in this new field.
Author Response
Dear Reviewers and Editor
The authors thank the reviewer very much for the comments. We have made a small integration within the introductory part of the text, inserting a small explanation of what libs is and what are the recent developments of the technique in the context of its use for automatic mineralogy.
The authors describe very well from abstract to conclusion the problematic of time consuming technics applied in mineralogy (especially with SEM-EDX...).
However, the authors are not referencing the LIBS technology applied for mineralogy (LIBS+AI=Automated mineralogy). This new technic is ultrafast for multielemental hyperspectral imaging of rocks, core and chips (including light elements and especially for critical minerals).
The authors should review the literature and report at least some papers in this new field.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Dear Authors and Editor,
this manuscript contains the description of an interesting new method, it clearly shows, how optical microscopy can better used to quantify some important metallurgical points. However, several questions are unclear and the manuscript was written in a bit of hurry. THat is why I suggest some revisions, written below.
Introduction
-line 40: 30 micrometer
-lines 50-51: AM is used not only for opaque minerals, maybe You should rephrase this sentence
-R% is actually depending on the wavelength of the light (that is why the opaque minerals have colour under reflected light), and it is not clear, how you treat this?
-please, emphasise, why it was useful to develop this new method? Is there any earlier method to do similar work?
Materials and methods
-line 101: there is a reference source error! here
-fig 1: the badly polished surfae is not affecting the results of the IA?
-there are four 1.1. chapters
-line 139: there is a reference source error! here
-line 154: there is a reference source error! here
-line 191: non opaque minerals will show the effects of transmitted light only with crossed polars, otherwise you are able to see its weak reflectance. Please, correct this.
-lines 191-192: internal reflection is NOT because of the proximity of opaque mineral. Please, correct this.
-lines 203 and 209 and 212: there is a reference source error here!
-line 224: what is this calibration curve here? Guess it should be the title of the next sub-chapter.
-it is not clear, with which software can you do this work? in the present form of this chapter the reader cannot use Your development, i.e. cannot work on his/her own sample to test Your method, because the practical description is not enough detailed.
SEM validation
-lines 264 and 267: there is a reference source error! here
-fig 7a: it would be better to show exaclty with arrows, where magnetite and goethite grains are located.
Results and discussion
-line 276: there is a reference source error! here
-there is no a and b parts of fig 8, please, check in the text, whether You are referencing to the right image
-you should include the optical microscopic images next to the SOM segmentatory image in fig 9, to be able to easier follow the statements. Also, plase include in the caption, what cyan, brown and blue, green are meaning.
-lines 326, 339: not ganga, but gangue
-you do not refer to figs 10, 11 in the text
-line 358: there is a reference source error! here
Conclusions
-this should be chapter 5
-it would be worth to mention, whether this method is possible to use in cases where more types of ore minerals present? wht would be needed for further development?
Author Response
Dear Reviewers and Editor
The authors thank you very much for the comments, corrections and questions that allowed us to reconsider some unclear aspects within the text. The corrections made are reported in this answer and in the red-colored text. As for the errors of the references in the text, they are probably due to a conversion error from word file to pdf file. However, before submitting the work, we will convert to static the references created with Mendeley, so that there can be no further problems.
Dear Authors and Editor,
this manuscript contains the description of an interesting new method, it clearly shows, how optical microscopy can better used to quantify some important metallurgical points. However, several questions are unclear and the manuscript was written in a bit of hurry. THat is why I suggest some revisions, written below.
Introduction
-line 40: 30 micrometer
the reference was corrected in micrometers
-lines 50-51: AM is used not only for opaque minerals, maybe You should rephrase this sentence
Thanks for the observation. We have rephrased the sentence
-R% is actually depending on the wavelength of the light (that is why the opaque minerals have colour under reflected light), and it is not clear, how you treat this?
We thank you for the observation. We did several tests, with different lighting sources, from d65 full spectrum to filtered light at + -546 nm and + -589 nm. We realized the variation in reflectivity, as reported by other authors. Finally, we chose to use the full spectrum, in the idea that not all have illuminators with monochromator or filter wheel.
-please, emphasise, why it was useful to develop this new method? Is there any earlier method to do similar work?
We have integrated the introductory text to point out the importance that this method could have compared to methods that use more complex IA such as CNN. Certainly, we reserve for the future to explore the part concerning the CNN, but not for a simple identification, but trying to extract and semantize information relating to textures, degree of release, orientation, and grain size. Currently, convolutional networks still have too high range of uncertainty, for example AlexNet has only 54% matching probability, this range can go up a little using Google's Deep Dream, but there are several orders of factors to consider such as magnitude of the 'iris and image resolution, size of the smallest detail that you want to try to discriminate, etc ...
Materials and methods
-line 101: there is a reference source error! here
-fig 1: the badly polished surfae is not affecting the results of the IA?
The SOM-type neural network, based on competitive algorithms, can be affected by poor sample polishing. The interesting point is that the problems related to bad polish are clustered in a separate segment, as well as the presence of residual noise due to different types of factors (bad white balance, noise correction, uneven lighting, etc.). This could be considered a bad thing, but the fact that these problems are separated from the artificial intelligence allows the operator, who will have to build a report, to evaluate them objectively and carry out the necessary corrections and procedures for a better evaluation of the sample. Artificial intelligence has enormous potential and even if unsupervised systems are used, a certain dose of heuristic correction, that is carried out based on the experience and knowledge of the operator, in our vision, is always indispensable. In our case, the samples were carefully polished and checked several times to put themselves in the most favorable analytical conditions. Furthermore, during the tests conducted we checked several times that the results that came out of the segmentation were not due to poor sample preparation or operator errors in the image acquisition phase.
-there are four 1.1. chapters
We have renumbered all the paragraph
-line 139: there is a reference source error! here
-line 154: there is a reference source error! here
-line 191: non opaque minerals will show the effects of transmitted light only with crossed polars, otherwise you are able to see its weak reflectance. Please, correct this.
We thank the reviewer for the observation. We have corrected the sentence.
-lines 191-192: internal reflection is NOT because of the proximity of opaque mineral. Please, correct this.
We thanks a lot for the observation, we have corrected with
-lines 203 and 209 and 212: there is a reference source error here!
-line 224: what is this calibration curve here? Guess it should be the title of the next sub-chapter.
We thank you for the observation. We have corrected the problem
-it is not clear, with which software can you do this work? in the present form of this chapter the reader cannot use Your development, i.e. cannot work on his/her own sample to test Your method, because the practical description is not enough detailed.
We added the following sentence to the paragraph:
For the analysis, a live script in Matlab environment has been developed. It combines the functions of some proprietary packages (ANN Tool) with original code written within our research group. The script is divided into four parts: i) acquisition and balancing of images in reflected and transmitted light; ii) enlargement of the data; iii) spatial reduction of segmentation; iv) production of outputs and data extraction from segments.
SEM validation
-lines 264 and 267: there is a reference source error! here
-fig 7a: it would be better to show exaclty with arrows, where magnetite and goethite grains are located.
We added the arrows as you suggested, Thanks a lot for the advise!
Results and discussion
-line 276: there is a reference source error! here
-there is no a and b parts of fig 8, please, check in the text, whether You are referencing to the right image
We have corrected the reference to Figure 10
-you should include the optical microscopic images next to the SOM segmentatory image in fig 9, to be able to easier follow the statements. Also, plase include in the caption, what cyan, brown and blue, green are meaning.
We corrected the figure by inserting the reflected light image on the side. We have also corrected the caption, indicating the phases to which the colors of the segments are associated.
-lines 326, 339: not ganga, but gangue
We corrected the Italian term "ganga" with the correct term "gangue"
-you do not refer to figs 10, 11 in the text
We have inserted references to figures 10 and 11 in the text
-line 358: there is a reference source error! here
Conclusions
-this should be chapter 5
We have renumbered all the paragraph
-it would be worth to mention, whether this method is possible to use in cases where more types of ore minerals present? wht would be needed for further development?
We started doing some experiments to verify the method's feasibility in the presence of several phases of opaque minerals. As a preliminary, we are observing that based on an optimization of the outputs of the neural network based on what we expect to find, a satisfactory result is obtained if the phases present show very different colors in reflected light. For phases with very similar colors, even when optimizing the number of output neurons, segmentation is sometimes not very effective. Probably. if you insert more neural networks in cascade (for example, the first that roughly segments the framed area and the second that segments within the area of the segments affected by phases of opaque minerals), you could succeed in resolving even imperceptible differences between the phases. Please notice that we still must verify it. Another way could be to semantize these phases according to aspects not exclusively related to the optical properties of reflected and transmitted light. In this case, it would become necessary to use convolutional networks instructed to recognize crystalline clothes and textures, structures, deformations, and so on.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Remove added text from line 454-458. The authors are self-citing several of their manuscripts as well they are not relevant for the current work and journal.
In the abstract line 23-24, authors are claiming a new, fast, and robust technics that can improve the lead time of quantitative automated mineralogy compare to SEM-EDS. The emerging LIBS technology must be separated in 2 categories. The first one is a traditional LIBS analysis speed (less than 20 Hz). The second is the new one that increase the speed up to kHz. This second category is more relevant to your proposed methodology for comparison. In another world, referenced from 35-46 must be reviewed.
36, 42 ultra low speed
35 Very low speed
37, 40 low speed
38 Review of LIBS technics (very interesting)
39, 41, 43 High speed
44, 45, 46 Not relevant (must be removed because they are not appropriate for minerals readers)
Author Response
Dear Editor and Reviewer,
The authors thank the reviewer for the careful work of rereading and criticizing the text which allowed us to review some considerations and focus our work in the best possible way. It also gave us the opportunity to consider further aspects to be taken into consideration for future developments.
Remove added text from line 454-458. The authors are self-citing several of their manuscripts as well they are not relevant for the current work and journal.
We thank the reviewer for the comment. We have removed the text from line 454 to 458 and relative bibliography
In the abstract line 23-24, authors are claiming a new, fast, and robust technics that can improve the lead time of quantitative automated mineralogy compare to SEM-EDS. The emerging LIBS technology must be separated in 2 categories. The first one is a traditional LIBS analysis speed (less than 20 Hz). The second is the new one that increase the speed up to kHz. This second category is more relevant to your proposed methodology for comparison. In another world, referenced from 35-46 must be reviewed.
36, 42 ultra low speed
35 Very low speed
37, 40 low speed
38 Review of LIBS technics (very interesting)
39, 41, 43 High speed
44, 45, 46 Not relevant (must be removed because they are not appropriate for minerals readers)
We thank the reviewer for the useful comment that was very useful for reorganizing the writing following a logic more closely linked to the nature of this work. We have therefore modified the text and the bibliography in a way that is more in keeping with the nature of the work
Author Response File: Author Response.pdf
