Pears Internal Quality Inspection Based on X-Ray Imaging and Multi-Criteria Decision Fusion Model

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This work presents the original and timely integration of X-ray imaging with an MCD-based classifier fusion for internal defect detection in pears. While individual components (e.g., DenseNet, SVM) are well-known, their integration under the MCD framework for this application is novel. This topic is highly relevant and aligns well with the scope of the Agriculture Journal. However, some concerns regarding your manuscript need to be addressed as follows:

 

1) The literature review could benefit from further references related to ensemble learning in defect detection and recent advancements in lightweight or explainable deep learning models applied to agricultural imaging.

2) No discussion is given on interpreting DCNN decisions (e.g., via Grad-CAM or SHAP). For practical deployment, this could be important.

3) No confidence intervals or hypothesis testing are provided for the model comparison results.

4) A discussion on latency, throughput, and potential integration into industrial sorting lines would significantly strengthen the application context.

5) While challenges with dataset size and confusion with the core/stem/calyx regions are acknowledged, there's little exploration of how these could be mitigated beyond the fusion model.

6) The dataset size still remains relatively small even after augmentation.

7) It is unclear how well the model generalizes to different pear varieties or unseen defect types.

8) Figures 11–14 are particularly useful, though adding captions that briefly describe the takeaways would improve reader comprehension.

9) To enhance clarity and emphasize the novelty of the research, it is advisable to refine certain expressions in the manuscript. Specifically, instead of using the phrase “internal defect detection in pears,” the term “non-destructive internal defect detection in pears” should be adopted, as it more accurately reflects the innovative nature of the approach and distinguishes it from conventional destructive methods. Furthermore, the term “handcrafted feature classifiers” could be replaced with “manual feature-based classifiers.” This substitution not only broadens the accessibility of the terminology for a wider audience but also aligns better with the commonly accepted nomenclature in machine learning literature. By incorporating these adjustments, the overall readability and impact of the text would be significantly improved.

Author Response

Comments 1: [The literature review could benefit from further references related to ensemble learning in defect detection and recent advancements in lightweight or explainable deep learning models applied to agricultural imaging.]
Response 1: [ Thank you for pointing this out. I agree with this comment. Therefore, I have added some literature on page five, in the last part of the Introduction.]

Comments 2: [No discussion is given on interpreting DCNN decisions (e.g., via Grad-CAM or SHAP). For practical deployment, this could be important.]
Response 2: [ Thank you for pointing this out. I agree with this comment. Therefore,  I added the relevant theories of DCNN in Section 2.3 on page 9.]

Comments 3: [No confidence intervals or hypothesis testing are provided for the model comparison results.]
Response 3: [ Thank you for pointing this out. I agree with this comment. Therefore, on page 22, I adopted a 1000-iteration bootstrap method to estimate the 95% confidence interval of each model.]

Comments 4: [A discussion on latency, throughput, and potential integration into industrial sorting lines would significantly strengthen the application context.]
Response 4: [ Thank you for pointing this out. I agree with this comment. Therefore, I added System Latency, Throughput, and Integration Feasibility in 4.2 on page 25.]

Comments 5: [While challenges with dataset size and confusion with the core/stem/calyx regions are acknowledged, there's little exploration of how these could be mitigated beyond the fusion model..]
Response 5: [ Thank you for pointing this out. I agree with this comment. Therefore, l conducted part of the discussion in the middle part of page 8.]

Comments 6: [The dataset size still remains relatively small even after augmentation.]
Response 6: [ Thank you for pointing this out. I agree with this comment. Therefore, I expanded on the original basis.]

Comments 7: [It is unclear how well the model generalizes to different pear varieties or unseen defect types.]
Response 7: [ Thank you for pointing this out. I agree with this comment. Therefore, I evaluated the generalization ability of this model on different pear varieties and unseen defect types in Section 4.1 on page 24 and conducted additional experiments.]

Comments 8: [Figures 11–14 are particularly useful, though adding captions that briefly describe the takeaways would improve reader comprehension.]
Response 8: [ Thank you for pointing this out. I agree with this comment. Therefore, I added captions that briefly describe the takeaways.]

Comments 9: [To enhance clarity and emphasize the novelty of the research, it is advisable to refine certain expressions in the manuscript. Specifically, instead of using the phrase “internal defect detection in pears,” the term “non-destructive internal defect detection in pears” should be adopted, as it more accurately reflects the innovative nature of the approach and distinguishes it from conventional destructive methods. Furthermore, the term “handcrafted feature classifiers” could be replaced with “manual feature-based classifiers.” This substitution not only broadens the accessibility of the terminology for a wider audience but also aligns better with the commonly accepted nomenclature in machine learning literature. By incorporating these adjustments, the overall readability and impact of the text would be significantly improved.]
Response 9: [ Thank you for pointing this out. I agree with this comment. Therefore, I made modifications in the text.l.]

 

Reviewer 2 Report

Comments and Suggestions for Authors

The study inspected Pears Internal Quality based on X-ray Imaging and Multi-Criteria Decision Fusion Model. The study seems interesting and important for the pear industry. Below are the comments.

1. Now a days most of these studies are undertaken using multispectral or hyperspectral imaging techniques, what is the advantage of using x-ray as detection tool over spectral techniques.
2. What is the generality of the study? Can this be used for other fruits and vegetables. What are the changes that need to be made in modeling to accommodate other fruits and vegetables.
3. What is the effect of x-ray imaging on quality parameters pears? Does it affect the chemical/nutrient composition of fruits?
4. How does this study help farmers/ industry/consumers? Are you developing any sorting machine using this study?
5. Please define the internal quality at one point in the manuscript to better clarify what are you inspecting exactly.
6. Paper has not followed the format of the journal. Please follow the standard format for writing manuscript which usually has Introduction, Materials and Methods, Results, discussion and conclusions. In this study, Materials and Methods, Results, discussion is written differently without following the format.
7. Citation style is not standard and needs to be changed as per journal format.
8. Why did you select DenseNet-121, LBP-SVM, and HOG-SVM classifier, give a proper justification of selection criteria.
9. What are the limitations of studying? Please mention that in the discussion.
10. Increase the number of references to a minimum of 30.
11. Several other comments are given in the attached pdf.

Comments for author File: Comments.pdf

Author Response

Comments 1: [Now a days most of these studies are undertaken using multispectral or hyperspectral imaging techniques, what is the advantage of using x-ray as detection tool over spectral techniques.]
Response 1: [ Thank you for pointing this out. I agree with this comment. Therefore, I explained this issue on the fourth page. In contrast, X-ray imaging is fast, non-destructive, and capable of detecting internal defects with high resolution.]

Comments 2: [What is the generality of the study? Can this be used for other fruits and vegetables. What are the changes that need to be made in modeling to accommodate other fruits and vegetables.]
Response 2: [ Thank you for pointing this out. I agree with this comment. Therefore,  On page 4.1 of 24, in order to evaluate the generalization ability of this model on different pear varieties and unseen defect types, I conducted additional experiments. Firstly, different types of pears were selected, along with some invisible defects. Future improvements may include incorporating more types of pear samples and defect types into the training dataset, as well as leveraging advanced techniques such as transfer learning and synthetic data generation.]

Comments 3: [What is the effect of x-ray imaging on quality parameters pears? Does it affect the chemical/nutrient composition of fruits?]
Response 3: [ Thank you for pointing this out. I agree with this comment. Therefore, I explained on pages 4 and 20 that before the formal experiment began, in order to verify whether X-ray imaging would affect the chemical and nutritional components of pears, we measured the sugar content, acidity and vitamin C levels before and after imaging. The results show that it will not affect its appearance, hardness, sugar content or other chemical parameters..]

Comments 4: [How does this study help farmers/ industry/consumers? Are you developing any sorting machine using this study?]
Response 4: [ Thank you for pointing this out.Yes, we are developing a pear grading device, which is described in 4.2 on page 25.]

Comments 5: [Please define the internal quality at one point in the manuscript to better clarify what are you inspecting exactly.]
Response 5: [ Thank you for pointing this out. I agree with this comment. Therefore, I define it in the first fragment on page 2.]

Comments 6: [Paper has not followed the format of the journal. Please follow the standard format for writing manuscript which usually has Introduction, Materials and Methods, Results, discussion and conclusions. In this study, Materials and Methods, Results, discussion is written differently without following the format.]
Response 6: [ Thank you for pointing this out. I agree with this comment. Therefore, I have made the revisions.]

Comments 7: [Citation style is not standard and needs to be changed as per journal format.]
Response 7: [ Thank you for pointing this out. I agree with this comment. Therefore, I have made the revisions.]

Comments 8: [Why did you select DenseNet-121, LBP-SVM, and HOG-SVM classifier, give a proper justification of selection criteria.]
Response 8: [ Thank you for pointing this out. I agree with this comment. Therefore, I explained it in the first paragraph on page 12.]

Comments 9: [What are the limitations of studying? Please mention that in the discussion.]
Response 9: [ Thank you for pointing this out. I agree with this comment. Therefore, I discussed in Section 4.1 on page 25 that there would be a decrease in the accuracy rate in the detection of different types of pears, and the effect would be average in the detection of non-obvious defects.]

Comments 10: [Increase the number of references to a minimum of 30.]
Response 10: [ Thank you for pointing this out. I agree with this comment. Therefore, I have made the revisions.]

Comments 11: [Several other comments are given in the attached pdf.]
Response 11: [ Thank you for pointing this out. I agree with this comment. Therefore, I have made the revisions.]

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

No more comments

Author Response

Thank you for your suggestion.