Predicting Quality Properties of Pears during Storage Using Hyper Spectral Imaging System

Round 1

Reviewer 1 Report

The manuscript describes a method for determining quality attributes of pears by means of Vis/NIR spectroscopy and chemometrics. Four parameters were considered: Soluble Solids Content (SSC), pH, Titratable Acidity (TA) and Vitamin C content.

The variation of such parameters depending on storage conditions (temperature, and storage time) was also studied, which is the most interesting part of the paper, because it leads the way to an on-line, nondestructive monitoring of fruit quality during storage.

The experimental design is sound, and the subject is clearly exposed. The manuscript is well provided with plots and tables as well as bibliographic references. However, there are same missing details and unclear expression that require a minor revision of the text before acceptance.

I have found no information about the dimension of dataset. How many pears were analyzed? How many fruits were allocated to each storage temperature? How many samples were used in training and how many in validation?

Were PLS models created using the whole spectrum or a suitable band was selected? I have seen considerable noise above 1050 nm in almost all spectra plots. Was such range employed on not in PLS calibration?

In the conclusion, at lines 235-241 outliers in the dataset are mentioned, but there is nothing about outlier in results discussion (chapter 3). It would be interesting to give some information about the percentage of removed outliers, at least for the most important models.

Line 138: The meaning of “exit-shift” cross-validation is unclear. It would be better to give information about how the training set was partitioned for cross-validation. For example: Leave-one-out, N-fold (where N is an integer), random or systematic sorting etc.

Line 186-187: The expression “no consistent trends in the spectral curves across different preprocessing approaches” is rather obscure. Please, rephrase it in a more straightforward way.

 

Author Response

A1. We noticed that the length of the present version is a little shorter than what we expected for the article paper (more than 4000 words suggested), counting from Abstract to Conclusion.

Number of words after revision is 6150 words now.

A2. To increase the readability of the article and to produce a deeper understanding of the research content for readers, we kindly suggest you add more details and references (more than 30 Refs suggested) to support your research results

 

In response to the esteemed reviewer's feedback, we recognized that the article's length did not meet the 4000-word criterion. Consequently, we undertook a thorough revision of specific sections, particularly the abstract, introduction, and portions of the methodology. All modifications have been highlighted in yellow for easy identification. It's crucial to emphasize that the core essence and concepts remain intact; we have merely enhanced the text with additional comments while preserving its original meaning.

 

A3) The manuscript describes a method for determining quality attributes of pears by means of Vis/NIR spectroscopy and chemometrics. Four parameters were considered: Soluble Solids Content (SSC), pH, Titratable Acidity (TA) and Vitamin C content.

The variation of such parameters depending on storage conditions (temperature, and storage time) was also studied, which is the most interesting part of the paper, because it leads the way to an on-line, nondestructive monitoring of fruit quality during storage.

The experimental design is sound, and the subject is clearly exposed. The manuscript is well provided with plots, and bibliographic references. However, there are same missing details and unclear expression that require a minor revision of the text before acceptance.

I have found no information about the dimension of dataset. How many pears were analyzed? How many fruits were allocated to each storage temperature? How many samples were used in training and how many in validation?

 

Thank you for your feedback. To address your queries on the dataset:

1. Number of Pears & Storage Allocation: Details are provided in the manuscript.

To ensure the consistency and integrity of our dataset, only pears that were free from any damage, shock, or disease were selected for the study. Following this, the pears were systematically grouped into 9 distinct sets, each containing 15 pears.

2. Training and Validation Samples: Distribution is outlined in the 2.4. Data pre-processing section.

A4) Were PLS models created using the whole spectrum or a suitable band was selected? I have seen considerable noise above 1050 nm in almost all spectra plots. Was such range employed on not in PLS calibration?

 

Thank you for pointing out the concerns regarding the spectral range used for PLS models.

 

To clarify, we observed noise at the beginning and end wavelengths of the spectrum. As a result, we removed these areas. The 465-1045 nm spectral range was subsequently used for analysis and modeling. This detail has been added to the manuscript for clarity.

 

A5) In the conclusion, at lines 235-241 outliers in the dataset are mentioned, but there is nothing about outlier in results discussion (chapter 3). It would be interesting to give some information about the percentage of removed outliers, at least for the most important models.

The following paragraph was added in the 2.4.1 section:

In the application of the PLSR method, specific data points may be classified as outliers. Such data, either inadequately represented by the model or exerting excessive influence, necessitate removal to ensure model accuracy (Wang et al., 2020). Analysis has indicated that samples with elevated residuals, despite their poor representation by the model, might not pose issues unless they have a significant influence, as characterized by Hotelling’s T² (Rafajłowicz et al., 2019). Critically, data exhibiting both high residual and Hotelling’s T² values emerge as particularly problematic, as they can dominate several model components and warrant their exclusion (Camacho et al., 2020). The SNV pre-processing method yielded no outliers, while D1 and D2 recorded the highest. Notably, the amalgamation of specific pre-processing methods, such as SG and MSC with D1 and D2, resulted in a substantial reduction in outlier incidence, likely due to the elimination of non-essential spectral regions. This intricate relationship is elucidated in Figure 5, which presents the residuals and influence of samples across varied pre-processing methods.

 

In Table 1, the number of identified outlier data in models built based on various pre-processing methods is presented.

Table 1: Outliers Detected Based on Pre-processing Method (F-Residuals represents the distance of the sample from the model. Hotelling’s T² measures how accurately the sample is described by the mode, and Both denotes samples that fall into both critical areas.)

 

Statistical criteria	Pre-processing method
	Raw	SG	SNV	Baseline	MSC	D1	D2	SG+ Baseline	SG+MSC+D1	SG+MSC+D2
F-Residuals	0	0	0	4	1	4	4	4	1	1
Hotelling’sT²	2	2	0	6	0	5	8	8	1	3
both	0	0	0	0	0	2	1	0	0	0

 

 

Figure 5: Comparison of Residuals and Influence Across Different Pre-processing Techniques

 

A6) Line 138: The meaning of “exit-shift” cross-validation is unclear. It would be better to give information about how the training set was partitioned for cross-validation. For example: Leave-one-out, N-fold (where N is an integer), random or systematic sorting etc.

Thank you for highlighting the ambiguity surrounding the term “exit-shift” cross-validation.

To address this, we used full cross-validation where only one sample is left out at a time, often referred to as the "Leave-one-out" method. This clarification has been added to the manuscript.

A7) Line 186-187: The expression “no consistent trends in the spectral curves across different preprocessing approaches” is rather obscure. Please, rephrase it in a more straightforward way.

 

 

The following paragraph was added in the 2.4.1 section:

"The superiority of the preprocessing technique S.G.+MSC+D2 over other preprocessing methods can be attributed to the extensive range of variations observed in the traits measured in this study. The spectral curves of various treatments do not follow a similar trajectory. Specifically, the spectral curves related to the 10-degree temperature treatment do not exhibit prominent peaks. Consequently, preprocessing methods focused on normalization and standardization can obliterate these curve peaks, resulting in the loss of valuable information. However, differentiation accentuates the latent information within the spectra, establishing a more precise correlation between radiation absorption and the targeted attributes (Grant and Bhattacharyya, 1985)."

 

 

 

 

 

 

Author Response File: Author Response.docx

Reviewer 2 Report

The study is about assessing pear fruit quality during storage. The structure of the paper is well organized. This paper is recommended for publishing but authors should take into consideration the following revisions:

·       Authors should avoid using undefined abbreviations in the abstract.

·       Why you choose only three temperatures (0.5, 5, and 10 °C)? is that enough to explore what is happening in this range of temperature?

·       The results discussion is poor, authors need to improve this aspect regarding the existing literature.  

Author Response

The study is about assessing pear fruit quality during storage. The structure of the paper is well organized. This paper is recommended for publishing but authors should take into consideration the following revisions:

Thank you for the constructive feedback and the recommendation for publishing. We appreciate your remarks and will ensure that the necessary revisions are made as per your suggestions. The authors would like to mention this point that:

In response to the esteemed reviewer's feedback, we recognized that the article's length did not meet the 4000-word criterion. Consequently, we undertook a thorough revision of specific sections, particularly the abstract, introduction, and portions of the methodology. All modifications have been highlighted in yellow for easy identification. It's crucial to emphasize that the core essence and concepts remain intact; we have merely enhanced the text with additional comments while preserving its original meaning.

 

 

B1) Authors should avoid using undefined abbreviations in the abstract.

 

Modifications have been made as suggested.

 

B2) Why did you choose only three temperatures (0.5, 5, and 10 °C)? Is that enough to explore what is happening in this temperature range?

 

As recommended, the ideal storage temperature for pear fruit is 0°C, allowing prolonged retention of quality attributes (Wang et al., 2014). Additionally, the temperatures of 5°C and 10°C were included, reflecting typical environmental temperatures in Ardabil during pear storage. This has been detailed in lines 91-95.

 

B3) The results discussion is poor; authors need to improve this aspect regarding the existing literature.

 

Thank you for your valuable feedback regarding the discussion of our results. We acknowledge the importance of placing our findings in the context of existing literature. We have made efforts to address this by revisiting and rewriting our conclusion. Additionally, throughout our results section, we have compared existing literature, as evidenced by statements concluding with "This has been previously reported..." It has been highlighted throughout the manuscript.

 

We aimed to present our results concisely to avoid redundancy, believing that repetitively referencing prior works might detract from the flow of the article. However, we understand the importance of thoroughness in academic discussions, and we are receptive to further suggestions or specific areas you believe could benefit from deeper contextualization within the literature.

 

Once again, thank you for your insights. We appreciate the opportunity to improve our manuscript and are open to any additional guidance you can offer.

 

 

Author Response File: Author Response.docx

Reviewer 3 Report

Materials and Methods

1. Line 86  Only pears with the same weight and size, same is not suitable, it’s better to use similar.

2. Sample preparation  In this section, the author should provide the total number of the pears used for the absorption spectrum measurement, and should also provide the information for the pears used for the chemical properties, meanwhile the authors should explain if the above pears belong to the same batch (first measure the absorption spectrum, then measure the chemical properties using the same pear), which is necessary for the experiment.

3. Figure 1  It’s better to number the pears from the left to the right in alphabetical order in Figure 1.

4. Figure 2   This figure is a little messy. The spectroradiometer could not be distinguished clearly from the figure. The author should provide a more professional figure about the instrument.

5. Data pre-processing  In this section, please provide the software used to preprocess the raw data.

6. Partial Least Squares Regression (PLSR)  In this section, please check carefully the range of SDR which is inconsistent with the range provided in the reference 16-18. At the same time, please provide the software used to perform the PLSR model.

Results and Discussion

1. Table 4  Please give the full name of the LVs and explain the detail about LVs in the part of Materials and Methods.

2. The explanation for the increase of pH is the conversion of acid degradation compounds and their possible esterification into corresponding esters in Line 193-194, while the interpretation for the decrease of TA is the respiratory process in fruits during storage leads to the conversion of organic acids into sugars in Line 214-215. Please confirm which one is the main cause.

3. Line 197-202  The best accuracy was obtained by the S.G.+MSC+D1 model, i.e. the combined preprocessing method, therefore the author should not only provide the superiority of the derivation proprocessing method, but also discuss the superiority of the combined preprocessing methods.

Conclusion

1. Line 243  LVs is short for latent variable.

2. Line 242-245  The author explain “The preprocessed spectrum using SG+MSC+D1 resulted in the highest accuracy for modelling SSC and pH, attributed to more hidden variables (LV) rather than other preprocessing methods. Meanwhile, the pre-processed spectrum using SG+MSC+D2 resulted in the highest precision for modelling vitamin C and TA due to the larger number of LV”, what’s the meaning of more hidden variables and the larger number of LV. In fact the preprocessed spectrum using SG+MSC+D1 for SSC and pH didn’t have the more LVs, and the pre-processed spectrum using SG+MSC+D2 for TA possessed the relative lower LVs. Please explain this.

I don't feel qualified to judge about the English language and style

Author Response

C1) Line 86 only pears with the same weight and size, same is not suitable, it’s better to use similar.

In response to your feedback, we've revisited and modified our discussion sections to ensure clarity and comprehensiveness.

The authors would like to mention this point that:

In response to the esteemed reviewer's feedback, we recognized that the article's length did not meet the 4000-word criterion. Consequently, we undertook a thorough revision of specific sections, particularly the abstract, introduction, and portions of the methodology. All modifications have been highlighted in yellow for easy identification. It's crucial to emphasize that the core essence and concepts remain intact; we have merely enhanced the text with additional comments while preserving its original meaning.

C2) Sample preparation  In this section, the author should provide the total number of the pears used for the absorption spectrum measurement, and should also provide the information for the pears used for the chemical properties, meanwhile the authors should explain if the above pears belong to the same batch (first measure the absorption spectrum, then measure the chemical properties using the same pear), which is necessary for the experiment.

Thank you for highlighting the need for clarity in the sample preparation section. We've addressed your concerns regarding the number of pears and their batch consistency in the revised manuscript. Both measurements were conducted on pears from the same batch to ensure data consistency. We hope this revision meets your expectations and are open to further suggestions.

 

C3) Figure 1, It’s better to number the pears from the left to the right in alphabetical order in Figure 1.

Done and Thanks.

 

C4) Figure 2   This figure is a little messy. The spectroradiometer could not be distinguished clearly from the figure. The author should provide a more professional figure about the instrument.

The spectroradiometer has been distinctly highlighted.

C5) Data pre-processing: In this section, please provide the software used to preprocess the raw data.

The software's name was cited in the revised article.

C6) Partial Least Squares Regression (PLSR)  In this section, please check carefully the range of SDR which is inconsistent with the range provided in the reference 16-18. At the same time, please provide the software used to perform the PLSR model.

Done and thanks.

C7) Table 4 Please give the full name of the LVs and explain the details about LVs in the part of Materials and Methods.

This was detailed in the revised article.

C8) The explanation for the increase of pH is the conversion of acid degradation compounds and their possible esterification into corresponding esters in Line 193-194, while the interpretation for the decrease of TA is the respiratory process in fruits during storage leads to the conversion of organic acids into sugars in Line 214-215. Please confirm which one is the main cause.

Thank you for pointing out the potential inconsistencies between lines 193-194 and lines 214-215. Both transformations (processes) mentioned indeed have closely related mechanisms and outcomes. It's challenging to definitively ascertain which process exerts a more pronounced effect on the pH increase or acid decrease. We believe that both processes collectively contribute to the observed changes, and isolating the dominant factor would require more intricate experimental designs.

We will make sure to clarify this point in the revised manuscript to ensure the readers have a comprehensive understanding.

C9) Line 197-202  The best accuracy was obtained by the S.G.+MSC+D1 model, i.e. the combined preprocessing method, the author should not only provide the superiority of the derivation preprocessing method but also discuss the superiority of the combined preprocessing methods.

We've clarified the matter in the revised manuscript.

C10) Line 243  LVs is short for latent variable.

Done.

C11) Line 242-245  The author explain “The preprocessed spectrum using SG+MSC+D1 resulted in the highest accuracy for modelling SSC and pH, attributed to more hidden variables (LV) rather than other preprocessing methods. Meanwhile, the pre-processed spectrum using SG+MSC+D2 resulted in the highest precision for modelling vitamin C and TA due to the larger number of LV”, what’s the meaning of more hidden variables and the larger number of LV. In fact the preprocessed spectrum using SG+MSC+D1 for SSC and pH didn’t have the more LVs, and the pre-processed spectrum using SG+MSC+D2 for TA possessed the relative lower LVs. Please explain this.

 

For clarity, we've added a detailed paragraph in the revised article.

 

"There exist no definitive guidelines or rationales upon which one might elucidate the pre-eminence of one preprocessing technique over another. The intrinsic characteristics of the raw spectral curves, coupled with the variance inherent in the output variable, lead to divergent scenarios concerning optimal preprocessing. For instance, an augmentation in the latent variable might either enhance or diminish a model's accuracy. Likewise, either individual or amalgamated preprocessing approaches could potentially bolster model precision. A perusal of extant literature reveals no discernible pattern in pinpointing the quintessential preprocessing method. Notably, in most studies[REF], a comprehensive array of preprocessing methodologies, as well as their combinations, have been employed for modeling purposes. The optimal preprocessing discerned in any given research is inherently bespoke to the dataset utilized therein and resists broad generalization to novel datasets. Moreover, the proximate accuracies achieved by all models render the identification of explicit reasons for the marginal superiority of a specific preprocessing technique over others a nuanced and intricate endeavor."