A Machine Vision-Based Algorithm for Color Classification of Recycled Wool Fabrics

Round 1

Reviewer 1 Report

General Comments:

In this paper, a machine vision combined with ANN is devised for carrying out a reliable color classification of regenerated plain colored wool fabrics. And demonstrating its effectiveness in comparison to other color classification approaches devised for textile and industrial fields. Overall, the subject is interesting and meaningful. The color parameters extraction processing method is relatively novel. But there are major issues in this article that need to be corrected and explained.

Specific comments:

1. The structure can be improved. This article lacks a conclusion.

Abstract:

1. In the Abstract, the introduction of the background is too much, but the significance of the work and work content is not enough. The abstract should highlight the innovation.

Introduction:

1. Line 71-95: There are several related researches are proposed in this paper, but there are no any results included of those researches. It is recommended to cite specific data from previous scholars to support your point of view, such as accuracy.

2. There are many color classification models based on machine vision and machine learning, such as support vector machine and extreme learning machine. Regarding the Introduction, I think readers would like to know more information about the advantages of ANN models. How about deep learning? I notice that the author also cites reference ([5]) about deep learning.

Materials and methods:

1. Line 164: “c is an integer in the range 1-10 representing the assigned color class”. The number of color class is 40. Is it “1-40” here? Not “1-10”.

Results and Discussion:

1. Line 309-310: The second table doesn’t have the title.

2. The description of Table 2 should be more detailed. The reader would like to know the meaning of the values in each column in Table 2, the author should give each column an appropriate column name. Some column names are same, such as “Family” and “Fabric id.”.

3. Line 295, “In the same table both the best (for fabric id. n1) and the worst (for fabric id. n130) …” In Table 2, the worst is fabric id. n143, not n130.

4. The authors should add a comparative experiment. It is suggested to compare with other models to explain the advantages of the model proposed in this study.

Conclusion:

The authors need to add conclusions.

Author Response

REV1

Dear Reviewer,

the authors express hearty thanks for providing additional suggestions for improving the scientific content of the manuscript. The concerns of the reviewers have been addressed to best possible way and detailed as follows.

 

General Comments:

In this paper, a machine vision combined with ANN is devised for carrying out a reliable color classification of regenerated plain colored wool fabrics. And demonstrating its effectiveness in comparison to other color classification approaches devised for textile and industrial fields. Overall, the subject is interesting and meaningful. The color parameters extraction processing method is relatively novel. But there are major issues in this article that need to be corrected and explained.

Answer: Thank you for appreciating our work. We have revised the paper according to your useful suggestions. Main changes are as follows:

- Abstract has been rewritten to be more concise and clear

- State of the art has been expanded and more works are considered, also with the aim of performing a more accurate comparison between our method and literature ones.

- Results are expanded to include the aforementioned comparison

- A new Section “conclusions” has been added

Specific comments:

1. The structure can be improved. This article lacks a conclusion.

As mentioned above we added this Section

Abstract:

1. In the Abstract, the introduction of the background is too much, but the significance of the work and work content is not enough. The abstract should highlight the innovation.

We changed the abstract to enhance the significance of our work

Introduction:

1. Line 71-95: There are several related researches are proposed in this paper, but there are no any results included of those researches. It is recommended to cite specific data from previous scholars to support your point of view, such as accuracy.

We expanded the SOA and we added also accuracy results from several other similar methods.

2. There are many color classification models based on machine vision and machine learning, such as support vector machine and extreme learning machine. Regarding the Introduction, I think readers would like to know more information about the advantages of ANN models. How about deep learning? I notice that the author also cites reference ([5]) about deep learning.

We also added more details on the usefulness of ANN and DL methods

Materials and methods:

1. Line 164: “c is an integer in the range 1-10 representing the assigned color class”. The number of color class is 40. Is it “1-40” here? Not “1-10”.

True, we corrected this.

Results and Discussion:

1. 1. Line 309-310: The second table doesn’t have the title.

Corrected

2. The description of Table 2 should be more detailed. The reader would like to know the meaning of the values in each column in Table 2, the author should give each column an appropriate column name. Some column names are same, such as “Family” and “Fabric id.”.

The description was wrong, sorry. In the new version, the column names are corrected and the description is clearer.

3. Line 295, “In the same table both the best (for fabric id. n1) and the worst (for fabric id. n130) …” In Table 2, the worst is fabric id. n143, not n130.

Corrected

4. The authors should add a comparative experiment. It is suggested to compare with other models to explain the advantages of the model proposed in this study.

As mentioned before, we added this part. In particular we added a new Table for this purpose.

Conclusion:

The authors need to add conclusions.

We added the Section

Reviewer 2 Report

The author used ANN for color classification of recycled wool fabrics. This paper has limited innovation. Some suggestions are as follows.

1. In the introduction, the literature summary is very thin.

2. This paper seems to adopt probabilistic neural network, but it also used a lot of "ANN". The use of these nouns may cause some misunderstandings.

3. Too little analysis and discussion of experimental results. More algorithms and evaluation parameters are recommended.

4. There are obvious formatting errors in the references. In addition, the number of references is too few. A large number of papers related to machine learning have been published in recent years.

Author Response

REV2

Dear Reviewer,

the authors express hearty thanks for providing additional suggestions for improving the scientific content of the manuscript. The concerns of the reviewers have been addressed to best possible way and detailed as follows.

The author used ANN for color classification of recycled wool fabrics. This paper has limited innovation. Some suggestions are as follows.

It is true that the innovation of our paper is not groundbreaking, but still the paper introduces a viable method for color classification which introduces simple devices and algorithms and that performs the classification with good results. Moreover, the classification is based on human know-how, and this is maybe the most relevant contribution of the paper. We hope this is more clear now.

We have revised the paper according to your useful suggestions. Main changes are as follows:

- Abstract has been rewritten to be more concise and clear

- State of the art has been expanded and more works are considered, also with the aim of performing a more accurate comparison between our method and literature ones.

- Results are expanded to include the aforementioned comparison

- A new Section “conclusions” has been added

 

1. In the introduction, the literature summary is very thin.

We considerably expanded the SOA and we added a lot of comments on the performance of similar methods, to compare our results with similar works in literature.

2. This paper seems to adopt probabilistic neural network, but it also used a lot of "ANN". The use of these nouns may cause some misunderstandings.

Correct, thank you! We changed ANN with PNN, it is more clear.

3. Too little analysis and discussion of experimental results. More algorithms and evaluation parameters are recommended.

We expanded discussion by adding the aforementioned comparison with previous works. In particular we added a new Table for this purpose.

4. There are obvious formatting errors in the references. In addition, the number of references is too few. A large number of papers related to machine learning have been published in recent years.

We corrected the format and we added 8 references

Reviewer 3 Report

The authors propose a machine vision system combined with a neural network based algorithm for performing color classification of regenerated plain colored wool fabrics. The solution is very interesting and important for the so called environmental sustainability, eco-design and circular economy. The paper is well writen, and some points deserve special attention in order to improve paper quality and readers' understanding.

 

"When tested against a dataset of fabrics, the created system enables automatic classification with a reliability index of about 93%, thus demonstrating its effectiveness in comparison to other color classification approaches devised for textile and industrial fields." -> please provide a deeper comparison between the related work and the proposed one. It would be nice to have a table making expliciting in which cases each solution performs well, and the limitations of each of them.

 

It seems the approach present in the paper is too simple. If it solves the problem and beats the state of the art, great! But I believe there are other works that should be mentioned and compared in the paper. For instance, "Defining a deep neural network ensemble for identifying fabric colors" is a recent paper that improved the state of the art on identification of fabric colors, so it would be important to have the proposed work compared to this one.

 

In order to simplify the problem, only the four most representative colors are selected from the image. It would be nice to see the influence of this number in the accuracy of the classification. How would that be?

 

Some general comments and minor errors are listed as follows.

 

""sustainable business,"" -> ""sustainable business","

"CCM" means Computer Color Matching

"provided in [13] where " -> "provided in [13], where "

"the main aim of the present paper" -> "the aim of the present paper"

"image processing algorithm" -> "image processing algorithms"

"As widely recognized [14] " -> "As widely recognized [14], "

" the main aim is to " -> " the aim is to "

"SOM" ?

"summation layer computes" -> "summation layer, computes"

"competitive layer selects" -> "competitive layer, selects"

"Since the main aim here " -> "Since the aim here "

"class 24) its" -> "class 24), its"

"the PNN classify" -> "the PNN classifies"

"Accordingly the" -> "Accordingly, the"

There is no caption for the table in page 9. Also, please use "." instead of "," as decimal separator.

"in [7] even" -> "in [7], even" 

"a confidence level of 0.5%." -> is this number really correct? this is a very low confidence level

Author Response

REV3

Dear Reviewer,

the authors express hearty thanks for providing additional suggestions for improving the scientific content of the manuscript. The concerns of the reviewers have been addressed to best possible way and detailed as follows.

The authors propose a machine vision system combined with a neural network based algorithm for performing color classification of regenerated plain colored wool fabrics. The solution is very interesting and important for the so called environmental sustainability, eco-design and circular economy. The paper is well written, and some points deserve special attention in order to improve paper quality and readers' understanding.

 Answer: Thank you for appreciating our work. We have revised the paper according to your useful suggestions. Main changes are as follows:

- Abstract has been rewritten to be more concise and clear

- State of the art has been expanded and more works are considered, also with the aim of performing a more accurate comparison between our method and literature ones.

- Results are expanded to include the aforementioned comparison

- A new Section “conclusions” has been added

"When tested against a dataset of fabrics, the created system enables automatic classification with a reliability index of about 93%, thus demonstrating its effectiveness in comparison to other color classification approaches devised for textile and industrial fields." -> please provide a deeper comparison between the related work and the proposed one. It would be nice to have a table making expliciting in which cases each solution performs well, and the limitations of each of them.

We added a number of results from similar works in literature, and we compared the performance of our method with the ones from previous works. In particular we added a new Table for this purpose.

 

It seems the approach present in the paper is too simple. If it solves the problem and beats the state of the art, great! But I believe there are other works that should be mentioned and compared in the paper. For instance, "Defining a deep neural network ensemble for identifying fabric colors" is a recent paper that improved the state of the art on identification of fabric colors, so it would be important to have the proposed work compared to this one.

 We “enhanced” the fact that our approach is simple (we said “simple yet effective”). We are aware that the approach does not beat the state of the art but rather is comparable with other methods in the SOA. The main advantage of our tool are that it does not require complex devices to acquire images and that the algorithm is quite simple. Furthermore, we think that the tool is useful for classifying according to human know-how.

In order to simplify the problem, only the four most representative colors are selected from the image. It would be nice to see the influence of this number in the accuracy of the classification. How would that be?

You are right, in the original version this part was too weak. We added a Figure to understand why we decided to use the first four RAL values and we also explained the reasons behind our choice.

 Some general comments and minor errors are listed as follows.

 ""sustainable business,"" -> ""sustainable business","

Corrected

"CCM" means Computer Color Matching

Corrected

"provided in [13] where " -> "provided in [13], where "

Corrected

"the main aim of the present paper" -> "the aim of the present paper"

Corrected

"image processing algorithm" -> "image processing algorithms"

Corrected

"As widely recognized [14] " -> "As widely recognized [14], "

Corrected

" the main aim is to " -> " the aim is to "

Corrected

"SOM" ?

Self Organizing Map… we added the definition

"summation layer computes" -> "summation layer, computes"

Corrected

"competitive layer selects" -> "competitive layer, selects"

Corrected

"Since the main aim here " -> "Since the aim here "

Corrected

"class 24) its" -> "class 24), its"

Corrected

"the PNN classify" -> "the PNN classifies"

Corrected

"Accordingly the" -> "Accordingly, the"

Corrected

There is no caption for the table in page 9. Also, please use "." instead of "," as decimal separator.

Corrected

"in [7] even" -> "in [7], even" 

Corrected

"a confidence level of 0.5%." -> is this number really correct? this is a very low confidence level

We made a mistake, we intended that the differently acquired fabrics led to the same classification with an ERROR of 0.5%

 

Round 2

Reviewer 1 Report

All of my previous concerns have been addressed properly. I do not have further concern.

Author Response

Thank you for your work, we were able to ameliorate our study thanks to your fruitful comments.

Reviewer 3 Report

Most of the issues with the paper were solved, congratulations. I believe there is still one issue to be solved before paper acceptance. It follows:

Table 3 compares the result of the proposed approach against other two, a "computer vision-based method" and a "deep learning-based method". Since the results of the proposed approach are inferior and the authors state that the computational demand for the other two algorithms are far superior, this should be described. For instance, if in order to obtain the classification performances of the other two it is necessary to have a supercomputer, and the proposed method works with less a less restrict hardware platform, this must be mentioned. This counts as an advantage of the proposed solution and for that reason it should be clear in the paper.

 

Some general comments and minor errors are listed as follows.

 

"also the aforementioned method make" -> "also the aforementioned method makes"

"is 93.34% thus " -> "is 93.34%, thus "

" the accuracy of similar methods based on the use of neural networks, lead to accuracy" -> please rewrite

"The method proposed in this work tend" -> "The method proposed in this work tends"

"[nuovo]" -> ?

Author Response

Thank you for your suggestions, it really helped to improve our paper.

In the current version we implemented your last suggestions.