Efficient Defect Detection of Rotating Goods under the Background of Intelligent Retail
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe article proposes an algorithm for the detection of destroying goods in dynamic visual vending machines, employing CNN, GCN networks, and an adaptive subsampling partition memory strategy to achieve high accuracy and robustness in detecting abnormal goods. However, there are some issues that need to be reconsidered, which are as follows:
1) The article lacks a detailed introduction to the related work on anomaly detection. There is also a lack of a brief introduction to the comparative methods in the experimental part.
2) The content from lines 112 to 117 is similar to that from lines 137 to 141. It is advised to merge these into one part.
3) The methodology section, particularly line 118, mentions using Grad-CAM to generate an Attention Map, but lacks an introduction to the Grad-CAM method. The activation function of the attention map in formula (2) uses 20 as the threshold. Please explain how this threshold value of 20 was determined.
4) There are issues with formulas, punctuation, and grammar in the article. For instance, there are two periods at the end of line 121.
5) It is suggested that mathematical symbols be used to represent the Attention Map in formula (3).
6) The meaning of "16-channel" in line 161 is unclear.
7) It is recommended that the entire text be carefully edited, with special attention to English grammar, word usage, and sentence structure.
8) The quality of some figures must be imrpoved, such as Figure 4.
9) The experimental section uses the AUROC metric to evaluate the model but lacks an introduction to AUROC.
10) Figure 6 lacks an introduction to the six subfigures it contains, and there is no identification of the model evaluation metrics in subfigures (a), (b), (c), and (d).
11) Figure 7's comparison of model performance is not intuitive enough. The authors can consider whether it could be replaced with bar charts or line graphs.
Comments on the Quality of English Language
Author Response
Please see the attached file:Author's Reply to the Review Report (Reviewer 1).pdf
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsIn this manuscript, the authors present a method for defect detection of rotating goods in smart vending machine. This research is specialized but not worthless, which endow this work has certain significance in the professional field of engineering practice. In general, the authors provide a solid experiment work and well-organized text content. There are several suggestions can help the authors further improve the quality of this manuscript.
1. The primary goals of this research are not clear. According to the presentation that the authors provide in sixth paragraph of the ‘introduction’ part, the issues of defect detection within smart vending machine include diverse products, complex background, computational and storage capacities. However, the main work of this research focuses on solving the problems of rotational robustness and operation efficiency within defect detection, without considering the other problems mentioned above. So, the authors should emphasize this point directly to make it clear to the reader.
2. The title needs to be retaken. Throughout the text can be found that the main characteristics of the proposed defect detection model are to acquire the rotation invariant feature and improve the operation efficiency. There are no function module or optimization strategy specially designed for few-shot learning except for a set of sub-experiment on a small sample. Therefore, it is not appropriate to view few-shot learning as keyword for title formation. The authors need to choose a more relevant title.
3. The limitation of existing method in obtaining rotationally invariant features are not well summarized and analyzed. To be specific, what’s the shortcoming of GNN in references? Why does the author propose the framework of CNN+GNN? What's the motivation? What gaps can be filled in the existing GNN approach? Does the GNN model adopted in the CNN+GNN framework contain an innovative part? or just employ the normal GNN model commonly used in the literature? The authors don’t explain well in manuscript.
4. The main innovation work needs to be well defined and summarized, which is beneficial to highlight the contribution and value of this manuscript. In my opinion, memory repository module construction and adaptive sampling are valuable innovations. Therefore, it is recommended that the authors summarize in this format at the end of the introduction to make it clear to the readers.
(1) ….
(2) ….
(3) ….
5. Why is the threshold used to define response levels in critical zones for enhancing signal detection set at 20? The authors do not provide a clear explanation.
6. The sample images in the defect detection flow chart should be changed from industrial parts to goods in smart vending machines to fit the topic of this research.
7. It is suggested to supplement schematics in section 2.2 to help readers better understand the technical details.
Comments on the Quality of English Language
English language is acceptable in general, but there are some grammatical errors that should be corrected. It should thoroughly proof read once.
Author Response
Please see the attached file:Author's Reply to the Review Report (Reviewer 2).pdf
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThis work presents an anomaly detection algorithm for object defect detection by images. The proposed Adaptive Sub-sampled Partition Memory (ASPM) model is well described and compared with several state-of-art algorithms. Except for several English grammar errors and PDF format issues, this paper is in good shape. Here are a few suggestions regarding this paper:
1,Figure, 2, 3, 4 are aligned to the left.
2, All math equation formats need improvements.
3, For Figure 7, it would be better to use a table instead of 3D picture to show the evaluation results of AUROS for different algorithms.
4, For Table 1, make sure the first character of each column name is uppercase ones.
Comments on the Quality of English LanguageThe English writing needs improvements. For example, the title of Figure 1 could be "The diagram of the ASPM model."
Author Response
Please see the attached file:Author's Reply to the Review Report (Reviewer 3).pdf
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThe authors have answered all my questions.
Comments on the Quality of English LanguageSome grammar mistakes still need to be carefully improved.
Reviewer 2 Report
Comments and Suggestions for AuthorsThe manuscript is well improved. Authors responded to all the queries raised at the time of the previous round of review. In my view, the present revised manuscript can be accepted for publication.
