Fabric Surface Defect Detection Using SE-SSDNet
Round 1
Reviewer 1 Report
I believe that the article "Fabric Surface Defect Detection Using SE-SSDNet" presents interesting research results but needs to be improved before publication.
- The abstract must provide readers with a critical and more detailed view of the document, in addition to that quantitative results must be shown.
- The introduction must be improved. It must be highlighted the originality of the research aims by demonstrating the need for investigations in the topic area. Please discuss the highlights individually and assure a clear correspondence between the affirmations from the manuscript and those from the cited papers. Please avoid bulk citations.
- Some figures are blurred to read, which should be avoided. I suggest improving the quality of the figures.
- The discussion section is missing. In the discussions section, a clear correspondence and comparison between the results of this study and those from the literature should be provided.
- The conclusion section is too short and can be improved in order to make the text more fluid and clearer to potential readers.
- An overall significant improvement in syntax and grammar is suggested.
- The template instructions were not followed.
- References should be prepared in accordance with the journal's recommendations.
- More than 50% of the reference articles are older than 5 years. The authors should reference recent literature.
- Use comprehensive and simple English sentences, there are some that are difficult to interpret.
Author Response
Point 1: The abstract must provide readers with a critical and more detailed view of the document, in addition to that quantitative results must be shown.
Response 1: We rewrote the abstract according to the structure and emphasis of the article
Point 2: The introduction must be improved. It must be highlighted the originality of the research aims by demonstrating the need for investigations in the topic area. Please discuss the highlights individually and assure a clear correspondence between the affirmations from the manuscript and those from the cited papers. Please avoid bulk citations.
Response 2: Thanks for the valuable comments provided by the reviewers, we rewrote the literature review and only wrote the description of the major related problems of the paper, which was divided into two parts at the same time, so that the readers could have a better understanding of the context of the article.。
Point 3: Some figures are blurred to read, which should be avoided. I suggest improving the quality of the figures.
Response 3: Thanks to the strict scientific attitude of the judges, we have redrawn some of the drawings
Point 4: The discussion section is missing. In the discussions section, a clear correspondence and comparison between the results of this study and those from the literature should be provided.
Response 4: Based on the reviewers' suggestions and the reasonableness of the article structure, we added a discussion section
Point 5: The conclusion section is too short and can be improved in order to make the text more fluid and clearer to potential readers.
Response 5: Thanks to the reviewer's suggestion, we have reconsidered the content of the conclusion and made changes
Point 6: An overall significant improvement in syntax and grammar is suggested.
Response 6: Thank you very much for the reviewer's suggestion, the grammatical problems of the whole article have been communicated with professional institutions and related personnel, and a lot of changes have been made.
Point 7: The template instructions were not followed.
Response 7: The article has been revised at the end according to the journal template
Point 8: References should be prepared in accordance with the journal's recommendations.
Response 8: Very much feeling the scientific attitude of the reviewers, we have combed through all the references and reworked and adjusted them
Point 9: More than 50% of the reference articles are older than 5 years. The authors should reference recent literature.
Response 9: Many thanks to the reviewers for their suggestions. Machine learning is evolving rapidly, and we have made adjustments to the references to remove some of the more dated literature.
Point 10: Use comprehensive and simple English sentences, there are some that are difficult to interpret
Response 10: Many thanks to the reviewers for their criticism. Throughout the article, we have made corrections in terms of language.
Author Response File: Author Response.docx
Reviewer 2 Report
The main contribution and proposed approach have some novelty in contribution. Revision in terms of technical details is needed before publication. So, some comments are suggested to describe technical details.
1. It is better to discuss about the runtime of your proposed approach briefly. It can be useful for real applications.
2. Report the input and output size of layers in the Figure 6.
3. How do you select the parameter alpha in the Eq. 5?
4. It is necessary to review more related papers about fabric defect detection such as machine learning-based methods. For example, I find a paper entitled “Fabric defect detection based on completed local quartet patterns and majority decision algorithm”, which has enough relation. As another example, I find a paper entitled “Multi-Resolution and Noise- Resistant Surface Defect Detection Approach Using New Version of Local Binary Patterns”. Cite these papers and some other related.
5. What is the difference between smooth function and popular activation functions?
6. Your proposed approach can be used in medical image analysis applications. For example, it can be used for non-healthy bacteria detection. I find a paper entitled “Isolation and characterization of lytic bacteriophages infecting Pseudomonas aeruginosa from sewage water”. Cite this paper and discuss about it as one of the advantages of your proposed approach briefly.
Author Response
Point 1: It is better to discuss about the runtime of your proposed approach briefly. It can be useful for real applications.
Response 1: For real-time consideration, the SSD model can be compressed by reducing the number of channels and setting a priori frame values that are more in line with the data according to the characteristics of cloth defects, especially the loss function can be optimized, and different model frames can be used for different textures of cloth. As the training time of SE-SSD is too long, the data set can be divided according to the proportion, 90% of the data set is di-vided into the training set and 10% of the data set is divided into the test set, and specific data enhancement means are applied to the training set to simulate various lighting conditions through color space transformation for data augmentation to improve the generalization ability and real-time performance of the model.
The running speed of SE-SSD in this paper is reduced to 57.3 fps even after optimizing the loss function, while the speed without Focal Loss function is reduced to 48.4 fps. However, compared with SSD, SE-SSD network increases mAP from 77.2 % to 78.6 % and 81.2 % to 82.0 % respectively. The results show that SE module can indeed improve the accuracy of model target recognition from the perspective of feature channel.
Point 2: Report the input and output size of layers in the Figure 6.
Response 2: Input image 300*300 when evaluating model,The network structure diagram is as follows
Overall detailed structure of the SE-SSD network
Point 3: How do you select the parameter alpha in the Eq. 5?
Response 3: alpha is the position loss weight, usually defaulted to 1, but if you want to improve the detection accuracy, the position accuracy requirements are not high, you can reduce alpha
Point 4: It is necessary to review more related papers about fabric defect detection such as machine learning-based methods. For example, I find a paper entitled “Fabric defect detection based on completed local quartet patterns and majority decision algorithm”, which has enough relation. As another example, I find a paper entitled “Multi-Resolution and Noise- Resistant Surface Defect Detection Approach Using New Version of Local Binary Patterns”. Cite these papers and some other related.
Response 4: In recent years, the use of fabric structure parameters for defect detection, fabric structure parameters mainly include fabric density, weave pattern, color pattern, etc., can be pre-set for these parameters raw, so that it can reduce the reliance on network parameters in defect detection, such as the transformation of LBP for face recognition, it can also provide some new ideas for other identification problems in the textile industry, such as fabric defect detection, fabric appearance analysis and fabric inversion modeling. The above two papers are very good and represent the next not research direction, but the identification of small defects is not very good
Point 5: What is the difference between smooth function and popular activation functions?
Response 5: The general image smoothing function is to remove the image texture and retain the larger edges in the image, and then the defective region and the background region as two different classes of pixel points, which can generally be used to segment the defective region and the background region using the K-means clustering method, and finally binarize the image so that the defects are completely separated; the activation function has a natural smoothing function, and its smoothing is carried out in accordance with the function prototype, which reveals the regularity of things more scientifically.
Point 6: Your proposed approach can be used in medical image analysis applications. For example, it can be used for non-healthy bacteria detection. I find a paper entitled “Isolation and characterization of lytic bacteriophages infecting Pseudomonas aeruginosa from sewage water”. Cite this paper and discuss about it as one of the advantages of your proposed approach briefly.
Response 6: Thanks to the scientific attitude of the reviewers, this is our mistake, SEnet is more active in medical image research, but we are not involved in medical images, we only take this method to use in textile images, we have not studied in depth how to use this method in medicine, and there is no relevant medical images, so in the revised version removed this aspect of the reference
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
Most of comments are considered in the revised version. The revised manuscript is better than original submission in terms paper organization and technical details.