An Objective Metallographic Analysis Approach Based on Advanced Image Processing Techniques
, Beñat Bravo 1, Iker Pastor-López 2 and Pablo G. Bringas 2
Round 1
Reviewer 1 Report
Comments for author File: 
Comments.pdf
Author Response
The paper titled "An objective metalographical analysis approach based on image processing advanced
techniques has a good research aim, the objectives are clear, and the conclusions are
convinced. However, after going through the whole paper, there are still a lot of improvements need
to be revised. Overall, the paper can be considered to publish with major revision.
1. The first seven paragraphs need to be re-written. You need to write so many paragraph
to talk about the history, a few concise lines should be good enough. As the research paper,
you should focus more on the relevant literature review studies, summarized the previous
work and find what is the gap that suitable for your study.
- First paragraphs have been rewritten. We still talks about general metallurgy and foundry (it is a way to focus the research) but more conciselly.
2. The references in your introduction part are too few, you should put more effort into the
introduction section.
- Due to we have rewritten the introduction section, we have also add new references.
3. Put more literature review of AI/Machine learning (different approaches: ANN/NN/SVM) in
the introduction section. Summarize what is done and what should be done in the future and
what is the key gap of AI/ML in the metallurgical field.
- Full introduction was rewritten, and we have take into account this comment to show possible solutions based on AI.
4. No need to describe the following outline in the last paragraph of INTRO, you need to conclude
what is the main content in your study.
analysis procedure, in section 3 we detail the results obtained from the experimentation that
we discuss in section 4 and, finally, in section 5 we sumarise the main conclusions and we
- We would like to mantain the distribution of the paper. We have improve the description to be more accurate in each section. Moreover, we have added an explanation of our main contributions in this paper.
5. Some of the literature review in 2.1 Nodule size section should rearrange to the 1. Introduction
section.
- Literature have been rearranged due to the introduction was more or less totally rewritten.
6. What is the unit of Table 4? % ?
- Yes, the unit is pectengae(%). We have added this information in both, header and caption.
7. How you define the good/regular/bad/pessimal? I couldn't find it, please define them in your Method section.
- A better explanation about the groups have been included in material and methods secton
8. I am very confused about the figure 5-7, are they are the machine learning results? You mentioned
the validation, is there any comparison could be shown as the graph?
- We have improve the explanation of the GradCAM in orther to improve the way of checking the DL method.
9. Please present more graphs in your work so that the readers can be more informed when reading the
paper and fetch the information easily.
10. The discussion and conclusion are okay, but could you make your short paragraphs into the long,
coherent paragraphs. In this way, it will be better to read. In the conclusion, make a [1,2,3,4] bullet
point, will be better to present.
- Authors have rewritten and reordered the conclusion section. In addition, we have added a summarisation in a enumeration way.
Reviewer 2 Report
This paper presents an image processing method to classify nodules from metalographies. It claims to address the following three research questions: (1) determining the size of the nodules; (2) determining the shape of the nodules; and (3) performing a characterization of the metallurgical quality.
The current paper lacks a clear description of the proposed approach and justification in experimental results.
1. The proposed approach is not described in detail. Section 2 needs to be significantly revised to present the proposed approach clearly, such as by using a flow chart.
2. The novelty of the proposed approach is not clear. It seems that only the existing methods are used in Section 2 to obtain various results. The deep learning method shown in Figure 3 is outdated.
3. Experimental results are very insufficient because there is no comparison with other methods.
4. The abstract needs to be revised to highlight the contribution of the proposed approach. Simply claiming that the proposed approach is automatic and objective is not convincing.
5. Section 3.3, the purpose of using GradCAM to explain the model results is not clear and not relevant to the focus of this paper.
6. The paper presentation needs to be checked carefully. For example, please adjust the alignment of line 131.
Author Response
This paper presents an image processing method to classify nodules from metalographies. It claims to address the following three research questions: (1) determining the size of the nodules; (2) determining the shape of the nodules; and (3) performing a characterization of the metallurgical quality.
The current paper lacks a clear description of the proposed approach and justification in experimental results.
1. The proposed approach is not described in detail. Section 2 needs to be significantly revised to present the proposed approach clearly, such as by using a flow chart.
- Intrduction was rewritten trying to plustrate the begining of the problems and also how authors face it to solve. Moreover, also section 2 have been rewwriten.
2. The novelty of the proposed approach is not clear. It seems that only the existing methods are used in Section 2 to obtain various results. The deep learning method shown in Figure 3 is outdated.
- More explanaitions added, better links and erason explanation. Totally rewritten the mayority of the paper.
3. Experimental results are very insufficient because there is no comparison with other methods.
- In tis case it is difficult to compare. We have added the comparison with the experts labels and the achieved reuslts. Al results were reorganised.
4. The abstract needs to be revised to highlight the contribution of the proposed approach. Simply claiming that the proposed approach is automatic and objective is not convincing.
- Abstract have been totally rewritten
5. Section 3.3, the purpose of using GradCAM to explain the model results is not clear and not relevant to the focus of this paper.
- Added some explanation to the use of GradCAM.
6. The paper presentation needs to be checked carefully. For example, please adjust the alignment of line 131.
- Presentations reviewed and adjustments done using hypenation command
Reviewer 3 Report
The authors presented a machine vision approach to analyze metallography. Various deep-learning algorithms were used to classify the shape of the nodules as well as their size of it. The authors also proposed their own deep learning-based algorithm for the same purpose.
1. Enormous information in the article discourages the readers and even sometimes misguides them. Almost every section is suffering from this problem. I would suggest removing very basic information. In the introduction section, paragraphs 2 to 8 look like a literature essay without any reference. Please remove/rewrite it.
2. Authors failed to mention the related article. Only two articles are included as the state-of-art article which is of course not enough. Without reviewing the latest articles, how can you propose the relevant research? I would suggest writing enough literate reviews mentioning all the related articles to date.
3. The important component of the research article is; background, related works, motivation of the work, and problem formulation. The authors completely forgot about the motivation of the research study. Some related articles have cited the materials and method section.
4. The methodology design does not reveal the novel idea of metallographic analysis. Section two also contains a lot of irrelevant information. Completely reformat the content and try to include graphical items including diagrams tables charts etc. which usually attract the readers and authors will be able to present the information effectively.
5. The presentation of the results is also so messy. The authors present a confusion matrix in a table which is not a common practice in a research paper. The fonts are also too large which dominates the overall figure. Please re-write all the confusion matrices or present the obtained results in a different way.
6. Implementation, limitation, and future direction of the research are not included properly.
7. The discussion section really does not discuss the idea authors presented, the outcome of the results, addressing the problem statement, etc. Actually, it does not connect the motivation of the research, experiments, and results. Need to improve the discussion section.
8. The conclusion does not reflect the actual conclusions drawn from the results. Re-write this section after improving the previous sections.
Author Response
The authors presented a machine vision approach to analyze metallography. Various deep-learning algorithms were used to classify the shape of the nodules as well as their size of it. The authors also proposed their own deep learning-based algorithm for the same purpose.
1. Enormous information in the article discourages the readers and even sometimes misguides them. Almost every section is suffering from this problem. I would suggest removing very basic information. In the introduction section, paragraphs 2 to 8 look like a literature essay without any reference. Please remove/rewrite it.
- The whole paper have been revisited and the mayority have been reorganised end rewritten. Authors think that the new point of view could solve this comment.
2. Authors failed to mention the related article. Only two articles are included as the state-of-art article which is of course not enough. Without reviewing the latest articles, how can you propose the relevant research? I would suggest writing enough literate reviews mentioning all the related articles to date.
- The bibliography have been increased, adding more iterature examples.
3. The important component of the research article is; background, related works, motivation of the work, and problem formulation. The authors completely forgot about the motivation of the research study. Some related articles have cited the materials and method section.
- All de original idea have been included and explained in both sections (introudcción and material and methods). We hope that this new organisation covers the requirements.
4. The methodology design does not reveal the novel idea of metallographic analysis. Section two also contains a lot of irrelevant information. Completely reformat the content and try to include graphical items including diagrams tables charts etc. which usually attract the readers and authors will be able to present the information effectively.
- We have added the methodology explained and rewritten the material and methods section.
5. The presentation of the results is also so messy. The authors present a confusion matrix in a table which is not a common practice in a research paper. The fonts are also too large which dominates the overall figure. Please re-write all the confusion matrices or present the obtained results in a different way.
- We organise and rewritte the results, trying to given a better readable text.
6. Implementation, limitation, and future direction of the research are not included properly.
- This information have been included in the results section and in the conclusion section.
7. The discussion section really does not discuss the idea authors presented,
the outcome of the results,
addressing the problem statement, etc. Actually, it does not connect the motivation of the research,
experiments, and results.
Need to improve the discussion section.
- Reviewed the discussion.
8. The conclusion does not reflect the actual conclusions drawn from the results. Re-write this section after improving the previous sections.
- Authors have reorder and rewrite the conlusion section of the paper.
Round 2
Reviewer 1 Report
the paper can be accepted in the present form
Author Response
We have review and recheck english. Regarding the background in the introduction. We have selected a Divide and Conquer mathodology and, then, more references were added in the explanation of each challenge.
Reviewer 2 Report
This paper has been significantly revised. There is no further comment.
Author Response
Ok, nothing to change. We recheck english.
Reviewer 3 Report
Many of the issues are solved in the revised version however some minor issues still need to solve before proceeding with the publication.
Abstract
Line12-Do you mean, in order to solve this problem???
Line 13-Machine Vision itself includes deep learning.
Line 15 - Give a reference or briefly explain the dataset.
The method and the results are not mentioned in the abstract. Please mention briefly the methods you used.
In the rest of the sections, there are still some redundancies but considerable.
Line 154 - This is not the correct way of referencing.
The conclusion section is too long. Please shrink it.
Author Response
Abstract
Line12-Do you mean, in order to solve this problem???
- Yes, change dto In order to solve this problem.
Line 13-Machine Vision itself includes deep learning.
- Yes, but we use both, ancient machine vision methods for extracting information and the deep leaning. We have added this information.
"combining classical machine vision techniques for feature extraction and deep learning technologies for making classifications,"
Line 15 - Give a reference or briefly explain the dataset.
- The data set was extracted from Azterlan's differente projects, so, it is not a dataset that people can download. In this way, we have completed the sentence with that information.
" combining classical machine vision techniques for feature extraction and deep learning technologies for making classifications, "
The method and the results are not mentioned in the abstract. Please mention briefly the methods you used.
- Added
"The proposed approach concludes that these techniques, a classification under a pipeline of deep neural networks and the quality classification using a ANN classifier,"
Line 154 - This is not the correct way of referencing.
- This paper was written in Latex, so th etemplate put the referencing in that way. We do not do it.
The conclusion section is too long. Please shrink it.
- Reduced
- Rechecked english
