Defect Detection in Printed Circuit Boards Using You-Only-Look-Once Convolutional Neural Networks
and Jiann-Shing Shieh 1Round 1
Reviewer 1 Report
- Using CNN for defect detection in PCBs is clearly plausible, and the results presented are moderately interesting.
- A question that immediately arises is whether the PCB defect detection problem is truly so challenging as to require a deep learning solution. Would ordinary machine learning methods, say, multilayer neural networks or tree based algorithms (e.g. decision tree, random forest), or even older technologies (e.g. expert systems), be sufficient to deliver solutions just as accurate?
- Most serious weakness of the paper is the poor English. The language must be improved significantly for the paper to be acceptable.
Author Response
Reviewer 1:
- A question that immediately arises is whether the PCB defect detection problem is truly so challenging as to require a deep learning solution?
Answer: Many thanks for your suggestion. We have added a paragraph. Please see our revised manuscript at page 2 of lines 41-55.
- Would ordinary machine learning methods, say, multilayer neural networks or tree-based algorithms (e.g. decision tree, random forest), or even older technologies (e.g. expert systems), be sufficient to deliver solutions just as accurate?
Answer: Many thanks for your suggestion. We have added a paragraph. Please see our revised
manuscript at page 13 of lines 367-383. In order to make the above explanation plausible we have trained a vanilla version of CNN methods to serve as the baseline to be compared with YOLO and compared the result. It is discussed from page number 11(line 319) to page number 12(line 351).
- Most serious weakness of the paper is the poor English. The language must be improved significantly for the paper to be acceptable.
Answer: Many thanks for your suggestion. We have done extensive editing to the document to improve the English.
Reviewer 2 Report
This research proposed the use of YOLO for defect detection of PCB. Overall, the research is interesting. However, a few comments are listed in below.
(1) The research motivation is lack of. It is not clear why the research is needed. For this, authors should review more literature relevant to PCB defect detection.
(2) The use of YOLO needs to be compared with other methods. Is it simple classifier such as the nearest neighbour, boosting classifier can achieve the same performance?
(3) It is strongly encouraged for the methods to be compared with another vanilla version of CNN methods to serve as the baseline to be compared with YOLO.
(4) Please improve the Figure and caption. Please make the captions more descriptive.
Author Response
Reviewer 2:
- The research motivation is lack of. It is not clear why the research is needed. For this, authors should review more literature relevant to PCB defect detection.
Answer: Thank you very much for your advice, I have added it on page2 line 41 - 55 to describe why exactly this research is needed and how does it makes the difference. Also, we review more literature relevant to PCB defect detection on page 2 line 56-81 and adding 9 references at page 15.
- The use of YOLO needs to be compared with other methods. Is it simple classifier such as the nearest neighbor, boosting classifier can achieve the same performance?
Answer: Thank you very much for your advice, I have added two paragraphs on page13 lines 367 – 383.
- It is strongly encouraged for the methods to be compared with another vanilla version of CNN methods to serve as the baseline to be compared with YOLO.
Answer: Thank you very much for your suggestion we have done the experiment and the result are discussed and compared from page number 11(line 319) to page number 12 (line 351).
- Please improve the Figure and caption. Please make the captions more
descriptive.
Answer: Many thanks for your suggestion. We have done extensive editing to the document to improve the English.
Round 2
Reviewer 1 Report
The language requires major improvement to meet publication standard of journals.
Author Response
The language requires major improvement to meet publication standard of journals
Answer: Many thanks for your suggestions. We have sent to professional English check company (i.e., Wallace Academic Editing) to correct and polish our English. Please see the following certificate and our revised manuscript.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Authors have addressed my concerns expressed in previous review comments. As such, I recommend it for the consideration of publication.
Author Response
Reviewer 2: Authors have addressed my concerns expressed in previous review comments. As such, I recommend it for the consideration of publication.
Answer: Thanks.
Author Response File: Author Response.pdf
