Semantic Segmentation-Based Identification and Quantitative Analysis of Cross-Sectional Quality Features in Luzhou-Flavor Liquor Daqu

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Although the manuscript is technically sound and pertinent to industrial computer vision, it needs to be strengthened in terms of novelty justification, validation, and clarity before it can be accepted.

Although the abstract is clear, it could be condensed by omitting some methodological details.

The terms "industrial inspection" and "small-object segmentation" are too general.

There are no current food vision system references (2023–2025) in the introduction.

Explain the term "pizhang" to readers from other countries when it first appears.

For clarity, figure captions require more thorough explanations.

Some sentences are difficult to read due to grammatical errors and length.

For pixel-based measurements, include units or a discussion of real-world conversion.

Tables 2 and 3's formatting could be made more readable.

Talk about inference time and computational cost for real-world implementation.

Enhance the conversation about failure cases, such as weak texture areas.

For more robust benchmarking, incorporate cutting-edge models (such as Transformer-based segmentation).

Include an ablation study for the components of the loss function (Lově, Dice, WCE).

Offer an analysis of performance gains' statistical significance.

 For industrial applicability, take into account pixel-to-real scale calibration.

 Extend the conversation about how well class imbalance handling works.

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

This work presents an integration of deep learning models and traditional industrial quality control methods by framing the evaluation of Daqu from subjective sensory judgment to an objective, data-driven framework. By developing segmentation-measurement pipeline, the authors address the challenges of complex internal structures and class imbalance typical in biological materials. The proposed solution is scalable and cost-effective. Overall, I think this is a well performed study written in organized manuscript.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

Dear Author, The “Semantic Segmentation-Based Identification -------study” presents a good and practical computer-vision-based approach for evaluating fermentation starters.

• There is no Clear identification of a relevant industrial problem.
• Use of multiple CNN architectures (U-Net, U-Net++, U2-Net) is good but no co relation are provided a solid comparative framework.
• Development of a pixel-level annotated dataset are not clear
• The proposed mask-driven quantification module are not defined in manuscript.
• Strong performance metrics are need to defined.

• The dataset size must described more explicitly.
• Clarify annotation methodology is important in this manuscript.
• The  preprocessing steps are required for parameter settings.
• Statistical significance of testing is important.
• Include comparison with conventional  evaluation methods to highlight practical advantages.

 

Comments on the Quality of English Language

The manuscript required general english corrections.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

Comments and Suggestions for Authors

Comments and Suggestions for Authors

The article entitled “Semantic Segmentation-Based Identification and Quantitative Analysis of Cross-Sectional Quality Features in Luzhou-Flavor Liquor Daqu” is interesting; however, it is somewhat difficult to understand. The authors may need to provide additional information to make the article more accessible to readers.

 

Remarks:

-   Abstract: Line 7

…….three representative CNN-based segmentation architectures…..

### From the phrase “…three representative CNN-based segmentation architectures…”, you should write out the full term of CNN (Convolutional Neural Network) and include the abbreviation in parentheses at its first occurrence. This will help readers better understand the content.

 

-   Based on the information presented in the manuscript regarding the three convolutional neural network models—U-Net, U-Net++, and U2-Net—the authors should provide a clearer explanation of the differences among these three models.

 

-   Page 3/18

 

### The authors state “660 images” in the table title. What does the column labeled “images” mean in Table 1? The authors should clearly explain what each column represents.

 

-   Page 3/18, Line 100-101

……During acquisition, a high-resolution smart-phone camera was used as the imaging device, with automatic HDR and AI enhancement…..

### Regarding the sentence, “…During acquisition, a high-resolution smartphone camera was used as the imaging device, with automatic HDR and AI enhancement…”, the authors should specify the resolution of the smartphone camera used in the study, or at least indicate the minimum camera resolution required.

 

-   Page 11/18, Line 329-330; 337-338

…….In terms of quantitative evaluation (see Figure 4(a), and Table 2), U2-Net achieved

the best overall performance across multiple key metrics……….…..Therefore, the overall superiority of U2-Net remains statistically and practically significant.

### From the statement:

“In terms of quantitative evaluation (see Figure 4(a), and Table 2), U2-Net achieved the best overall performance across multiple key metrics… Therefore, the overall superiority of U2-Net remains statistically and practically significant.”

Do you mean that U2-Net is the best model? If this model is applied in practice, what specific advantages would it provide? This point should be further clarified and elaborated.

 

-   Page 13/18, Line 377

3.3. Extraction and Visualization of Key Morphological Parameters

Model Prediction Results

Figure 6. Visualization process for extracting quality features from Daqu cross-sections.

### Based on Section 3.3, Extraction and Visualization of Key Morphological Parameters – Model Prediction Results, and Figure 6 (Visualization process for extracting quality features from Daqu cross-sections), the experimental results would be significantly strengthened if a comparison were provided between the model prediction results and classical methods. Such a comparison would help clearly demonstrate the advantages and improved accuracy of your model relative to traditional approaches, which heavily rely on expert knowledge and extensive practical experience.

 

-   Page 15/18,

4. Conclusion and Discussions

### This research has potential practical value; however, the authors should further elaborate on its significance and practical implications. In particular, they should clarify how the findings can be applied to the production, quality control, or evaluation of Luzhou-flavor Daqu, and expand the Discussion section accordingly.

 

### In the Discussion section, the following points should be further elaborated, particularly in relation to how the proposed model outperforms conventional methods:

• How does the proposed model compare with traditional evaluation approaches, and what specific advantages does it demonstrate?
• How does the improved accuracy influence the classification and grading of Daqu?
• Does the model contribute to reducing inspection time or operational costs?
• In what ways does it support smart fermentation or intelligent manufacturing systems?
• Does the framework have the potential to be extended to other types of Baijiu?

 

Cheers,

Date of this review

6 April 2026

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Accept.

Author Response

Thank you very much for your positive decision and for your careful handling of our manuscript. We sincerely appreciate your time and support, and we are very pleased that the manuscript has been accepted for publication. We have also refined the figures in the final revised version to improve readability and presentation quality.