From Harvest to Market: Non-Destructive Bruise Detection in Kiwifruit Using Convolutional Neural Networks and Hyperspectral Imaging

Round 1

Reviewer 1 Report

This study aims at an early, automatic and nondestructive detection and classification of bruises in kiwifruit based on local spatio-spectral near-infrared (NIR) hyperspectral (HSI) imaging. Overall, it was easier to detect bruises for unripen fruit than the ripen fruit, with higher correct classification rates for the former than the latter. Their results implied that early detection of the bruising area on HSI imaging was consistently more accurate in the unripen fruit compared to the ripen fruit.

 

I have some questions as follows:

 

What is the reason for selecting 8h and 16h for testing in this study?

 

L130- 420 kiwifruit were collected but why only 210 were used for experiment? 

 

Table 1- For training, would 68 fruit be enough to generate a reliable model?

 

Figure 3- This figure is of poor resolution and not self-explanatory. The authors should improve its quality first and then provide more information in the figure, including wave-length, classification, stages, etc.

 

L185- should be Fig. 4.

 

Figures 4 and 5- These two figures basically have no big differences and can be combined into one figure.

 

L430- An explanation of this fact might be the higher contrast of the color change after bruising of the fruit flesh in unripen as compared to ripen fruit, despite this hypothesis need to be further investigated for proper validation. 

 

How would the authors design experiments to validate this hypothesis?

 

L444- For next step, would this model be optimized and extended to apply for other fruits? And how much does this technology cost to implement?

 

Table 10- The table title says “The correct classification rates (CCR) are comparable”, but in the footnote it says “a direct CCR comparison is not possible since datasets are not common”. It is confusing and the authors need to further explain.

I feel that moderate editing of English language is required.

Author Response

Dear reviewer
Please see the attachment.
Best Regards

Author Response File: Author Response.pdf

Reviewer 2 Report

see the attachement 

Comments for author File: Comments.pdf

Moderate editing of English language required

Author Response

Dear reviewer
Please see the attachment.
Best Regards

Author Response File: Author Response.pdf

Reviewer 3 Report

The idea of the article is good but the presentation is a mess. The potential of hyperspectral imaging is not used or shown.

Figure 2. Unclear picture. Take a better picture or make a schematic drawing.

Figure 3. Uncomplete explanation. What are the rows and columns? Which wavelengths are used?

Figure 4 and 5. The background makes the text unreadable.

Figure 6. Mention which wavelength is used in 6C.

From Table 2 and Figure 7. Too many figures and tables that are just confusing.

Author Response

Dear reviewer
Please see the attachment.
Best Regards

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The author did not answer all the comments. not only did they not answer but actually deleted them from the response file. 

- author has not done any model section process. 

- I don’t see any contribution in models they have chosen. They have used pertained models without any modification.

- Table 2: It is requested that the author should provide a confusion matrix generated from testing but not manually created in Word/excel.

- No future directions are provided in the conclusion.

- The practical usage of the proposed model is not provided neither in the introduction nor in the conclusion.

Minor editing 

Author Response

Dear Reviewer
Please see the attachment
Best Regards

Author Response File: Author Response.pdf

Round 3

Reviewer 2 Report

Author has incorporated all the comments in second revision 

 Minor editing of English language required

Author Response

Dear Reviewer

Please see the attachment.

Author Response File: Author Response.docx