Detection of Bad Stapled Nails in Wooden Packages

Round 1

Reviewer 1 Report

Please consider following comments:

1) motivation section doesn't have any references, please add

2) on page 3 section 1.2, first three paragraphs can be merged as they talk about the same topic. Please avoid redundant paragraphing

3) I understand that the test was performed using a lab-scale sample, however it is not clear how many samples were used? was it just one crate?

4) the representation of confidence matric is odd. Please show it as a diagram. Is the confidence matrx same as confusion matrix?

5) Please draw most important conclusions and list them at the end of the article.

6) more description of the figure 6 is required.

7) overall, it looks like a simple classification problem and I fail to understand the novelty and merit of the approach. is the novelty in the application?

Please support the work addressing above comments.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

This article proposes a production line implementation scheme based on a conveyor belt automatic detection system, which can effectively detect the stitching of nails at box stitching and reduce item damage

However, there are some concerns as following:

1.  The writing can be more concise and to the point.

2. The article uses the method of projecting straight lines to determine the condition of nails. If the nails are well sewn, they will be projected as a straight line. If the nails are poorly sewn, the projected straight line will be broken. However, Figure 2 in the text only shows one situation, without a comparison between good and poor nail stitching. Moreover, we do not know what situation Figure 2 shows, as the ROI area light in the image is also discontinuous.

 

3. There are still many real-world scenarios for detecting sharp objects. This article only detects whether the nails at the seam of the box are well sewn, and the work is limited by the items and the scene, resulting in poor application mobility

4. In order to preserve the pixel features at the stitching point, it is necessary to ensure the relative positions of the camera, stitching box, and lighting. This places higher demands on the scene conditions, otherwise it will affect the experimental results and lack good adaptability.

5. In the work of this article, convolutional networks are used for classification and detection, and the transformer has achieved good results in the field of CV. Do you consider introducing the transformer into this work

6. Lack of some references about CNN and attentions, such as:

[1] Cross Channel Aggregation Similarity Network for Salient Object Detection, JMLC

[2] CSART: Channel and Spatial Attention-Guided Residual Learning for Real-Time Object Tracking, Neurocomputing

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

In this paper authors have developed a model “Detection of bad stapled nails in wooden packages.” The topic has some significant research value, however, the manuscript lacks the novelty which really holds this paper back. Following are my remarks.

1.The abstract seems vague, normally abstracts should contain problem statement, followed by the method description, results analysis and conclusive remarks.

2.The keywords are too general which are failed to highlight the main contribution

3. What is the main contribution of this work? It seems that authors have utilized the existing technologies. Main contribution should be discussed at the end of introduction section.

4. What was the motivation behind developing this system for in-line inspection of bad stapled nails in stitched crates?

5.  Could you elaborate more on the computer vision system used for identifying fails?

6. What were the image processing algorithms used in the developed inspection system?

7.  What were the peculiarities and application issues that you encountered while developing this system?

8. Could you describe the conveyor belt-based implementation of the inspection system for automatic inspection?

9. How did you achieve an accuracy of over 95%? What was the user-defined method used for defining nail location in the image?

1. What were the comparative results of the performance of the 5 state-of-the-art classification algorithms based on a deep neural network, and what were the trade-offs between speed and precision in the detection?

1.  How did you determine the definitive configurations of the inspection system, and what camera and light source positions did you test?

1.What are the potential applications of deep learning computer vision algorithms in realistic settings beyond this benchmark work?

1.The authors should compare their method with the SOTA methods to same objective.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The authors have addressed most of my concerns. 

Please revise and improve the paper before publication.

No other suggestions.

Author Response

The paper has been improved and revised according with the academic editor instructions.

Reviewer 3 Report

Authors have addressed all the issues raised in previous review , now the quality of manuscript seems better, there I have no objection for accepting this paper. Congratulations to Author 

Author Response

Thanks so much