Classification of Adulterated Particle Images in Coconut Oil Using Deep Learning Approaches
Round 1
Reviewer 1 Report
The paper is very interesting, well written and well structured.The methodological rigor with which the problem was faced is more than
satisfactory. The methodology presented is clear and the level to which
it has been detailed is remarkable.
I only ask that Section 3.3. sold described in more detail.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
After read the paper for version two, I have some suggestions:
1.Abstract:
(a)The keywords are still not matching with the contents, such as classification.
(b)The main focus is turbidity for oil, but in the abstract, it did not appear.
(c) Please adjust the rank of keywords, coconut oil should appear in the first.
Please recheck.
2.Please try to use third-person writing in the paper as possible as.
For example, page 1, line 20: We collect two…, May be can changed into The paper…., page 14, line 496, We…
3.Figure 3 already modified.
4.I still have a question. Does the entire verification process belong to the scope of prediction?
5.The conclusion already modified.
6.I think REF[9], REF[10] and REF[11] are the Bible in this field, if the author could, is there a new reference year?
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
I accept the paper without comments
Author Response
Thank you for all comments.
Reviewer 4 Report
This paper applies several famous neural networks directly from the references to the collected coconut oil datasets, PiCO_V1 and PiCO_V2, for impurity identification.
- Since the main contribution of this paper is the dataset, I think a link to the collected dataset should be shared in the manuscript.
- Besides the dataset, I think the contribution of this paper is very limited.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
Round 1
Reviewer 1 Report
*) The figures shown in the paper are very interesting. Then, it would be advisable for each of them to have a self-explanatory caption.
*) Formulas (1), (2), (3) and (4), although consolidated in the literature, are rather dated. Then, it would be appropriate to justify their use with respect to more recent formulations that would offer more performing results.
*) Explain more clearly why in Table 5 the numerical values of the main diagonal have been highlighted in bold.
*) As highlighted in Table 1, the images to be analyzed could be affected by uncertainties. Then, one might think of using a fuzzy preprocessing to improve its quality. Obviously, this use goes beyond the already valuable work done by the authors. However, I advise them to insert a sentence in the text that highlights this need, putting the following relevant works in the bibliography:
doi: 10.1109/ICSIPA.2015.7412240
doi: 10.1007/s40708-016-0045-3
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
The paper use ten methods in CNN to analyze the objects, and compare the final result by using they accuracy, it is a good approach in this field.
After read the paper, I have some suggestions:
1.Abstract: The keywords are not matching with the contents: classification; computer vision. Please recheck.
2.Please try to use third-person writing in the paper: For example, page 2, line 47: This paper… The paper, also in line 57 and line 63… . Same everywhere else
3.Please adjust the scale of Figure 3.
4.The equation shows it is by using algebra method. According to my personal opinion, it is based on pattern recognition, not prediction, it may be questionable.
5.The conclusion seems to be missing the limitations of the paper. Please recheck.
6.The REF[5], REF[8] and REF[9] are too far away. Especially, REF[5] is Conference paper, REF[8] is book, and REF[9] is the Bible in the field? Please recheck.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
The use of artificial neural networks to solve the presented research problem is not new. There are a number of publications that deal with the application of neural networks in general. These fragments (intoruction and literature review) of the publication should be supplemented with an in-depth review of the literature and a clear indication of the advantages of using convolutional neural networks over the solutions published so far.
Please also refer to the adopted accuracy of the network forecast (testing) at the level of 82%. What is the justification for such a value as acceptablee?
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments about the manuscript:
The fundamental and most important question, which is unanswered - why it is important to solve classification task? Even if there is some explanation for this, reading the article gives the impression that it is a binary task: oil with impurities, without impurities. Maybe it can be divided into four classes. As I understand, the first group - air, second group - fibers, third group - dust and the last one group - coconut tissue. However, there must be a clear reason for the exclusion of 10 classes
There is too little information about the coconut oil problem.
The analysis of the literature presents a classical theory that does not need to be repeated. It is enough simply put a link to CNN architectures. And what is CNN, deep learning or GoogleNet really doesn’t need to be written.
What is state of the art?
Summarizing, several classical CNN architectures taken for the task, providing classification accuracy, F1.score. The process of augmentation is not defined (what kind of augmentation) besides, this is done automatically for certain CNN architectures.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
The supplements to the literature review are not fully satisfactory.
Reviewer 4 Report
Thank you for the answers.
It is still unclear what ir the purpose of the classification into 10 classes. The answer that the division into classes of 4 expressions leads to major errors sounds questionable. "we initially used experts to classify objects into 4 groups: air, fiber, dust, and tissue. It was found that the diversity of shapes in each group led to high classification errors. "
There is a lack of information on the consequences of prediction errors for such classes.
The novelty of this work is too weak, as there are many such similar solutions from a technical point of view in other areas. Actually, there is no novelty from either the technical or the fundamental side.
