A Back Propagation Neural Network Model for Postharvest Blueberry Shelf-Life Prediction Based on Feature Selection and Dung Beetle Optimizer

Round 1

Reviewer 1 Report

The authors propose a theoretical basis for the shelf-life determination of blueberries under different storage temperatures and offered technical support for the prediction of remaining shelf life.

The experimental design is quite straightforward, and it is well-prepared to understand.

It was easy to follow the mainstream of the study.

The following points should be corrected for publication.

-        It is suggested to modify the title so that it does not contain abbreviations.

-        It is suggested to provide keywords that are not present in the title.

-        Given that shelf life is a complex trait highly influenced by genotype/variety and the environment, the authors are encouraged to discuss the utility of the proposed model for evaluating other varieties/genotypes and, if possible, extending its use for predicting shelf life in other crops. Perhaps testing the proposed model on multiple genotypes and in other crops could be a way to validate it.

-        In Figure 5, the captions are very small and hinder the interpretation of the graphs.

Author Response

Qusetion1: It is suggested to modify the title so that it does not contain abbreviations.

Answer1: Thank you to the reviewers for their valuable comments and suggestions. We strongly agree with you and have changed the title to A Back Propagation Neural Network Model for Postharvest Blueberry Shelf-Life Prediction Based on Feature Selection and Dung Beetle Optimizer.

Qusetion2: It is suggested to provide keywords that are not present in the title.

Answer2: Thank you to the reviewers for their valuable comments and suggestions. We very much agree with your views and have changed the keywords accordingly.

Qusetion3: Given that shelf life is a complex trait highly influenced by genotype/variety and the environment, the authors are encouraged to discuss the utility of the proposed model for evaluating other varieties/genotypes and, if possible, extending its use for predicting shelf life in other crops. Perhaps testing the proposed model on multiple genotypes and in other crops could be a way to validate it.

Answer3: We thank the reviewers for their valuable comments and suggestions on our study. We fully agree with the reviewers that shelf life is a complex trait that is strongly influenced by genotype and variety. Although the model proposed in this paper achieved high prediction accuracy at three temperatures, 0, 4 and 25 °C, we also recognise some limitations and shortcomings of our study, such as the use of only a single Liberty variety of blueberries as the subject of the study, which may be subject to varietal bias, and the lack of consideration of shelf-life prediction of other crops (e.g., fruits, vegetables, etc.). To address these limitations and shortcomings, we give several future perspectives at the conclusion as follows:
- In future studies, we consider training and testing our proposed shelf-life prediction model on multiple varieties or genotypes of blueberries to validate its utility in evaluating other varieties or genotypes and to compare it with other models or methods in terms of prediction effectiveness. By doing so, we hope to increase the generalisability and replicability of our findings.
- In our future research, we plan to extend our proposed shelf-life prediction model for predicting shelf-life of other crops (e.g., fruits, vegetables, etc.) and explore its similarities, differences, and patterns among different crops. By doing so, we hope to add innovation and value to our research.

Qusetion4: In Figure 5, the captions are very small and hinder the interpretation of the graphs.

Answer4: Thank you to the reviewers for their valuable comments and suggestions. We very much agree with your views and have changed Figure 5 accordingly.

Reviewer 2 Report

Dear Authors,the paper you have submitted is very interesting and brings many important aspects to the subject of storage and transport of blueberries. Please find below some questions and doubts, as well as suggestions for improvement of the paper.

The authors have attempted to build a Shelf-Life Prediction Model. For this purpose, they attempted to build a model that could predict the shelf-life of blueberries with high accuracy and at different temperatures. They meticulously addressed many problems regarding the process of modeling. They conducted a lot of tests and compared different algorithms, which enabled them to find the optimal solution to the problem.

2. Materials and Methods
2.1. Materials and Experimental Programme

How many times was the experiment repeated, and on the basis of how many fruits were conclusions drawn about the entire population? Can the conclusions be interpreted for all blueberries or only for a given variety? The samples seem to be too small.

How long was the storage time of the blueberries and what were the storage conditions from the moment of harvest to the start of the research (T0)? Under what conditions were the berries transported to the laboratory?

2.2.10. Sensory Evaluation Scores
The number of people who choose to conduct sensory assessment is too small to be convincing. What ages, sex and segments were the panelists?

Figures 5., 12., 13., 14. - must be improved, the graphs are too small and illegible

Tables 8., 9. - must be improved, there is a lack of data in the paper (?) may be you should give them in horizontal layout

lines 676-677: "This suggests that low temperature storage can delay the ageing process of blueberries by preserving anthocyanins, and that 0°C storage is more effective than 4°C." - this is obvious, your experiment does not bring anything new on this topic.

6. Discussion - there is the lack of propre discussion of the results, the discussion is not correct, in this chapter the authors should carry out a comparative analysis of their research results with the research of other authors, this chapter needs to be improved. Maybe you could discuss the obtained results to other papers discussing the quality of blueberries “Liberty” and the changes during storage. This could improve this section and improve the reliability of the proposed prediction model.

7. Conclusions

lines 832-838 are not conclusions but statements, in this section the results shouldn't be given, lines 836-838 - this is known and obvious.

Author Response

Q1: How many times was the experiment repeated, and on the basis of how many fruits were conclusions drawn about the entire population? Can the conclusions be interpreted for all blueberries or only for a given variety? The samples seem to be too small.

A1: We thank the reviewers for their valuable comments and suggestions on our study. We very much agree with your views. Firstly, for your question about the number of experimental samples and the number of experimental repetitions, we have given the corresponding descriptions about the number of experimental samples and the number of experimental repetitions in 2.1 and 2.2, respectively. As can be seen from 2.1, about 90 blueberries were used as experimental samples per day at each temperature (three temperature levels in total, that is, about 270 blueberries per day were used as experimental samples, totalling about 1890 blueberries). Since the experiments were conducted sequentially with sensory evaluation, TPA testing, and physicochemical index determination experiments (the indexes for destructive testing were fewer and at the end), we believe that the conclusions based on the approximately 1,890 blueberry fruits are reliable. We have checked and added some quality indicators in 2.2 where the number of experimental replicates was not written.
Secondly to your question about whether the conclusions of this paper can be interpreted as applying to all blueberries or only to specific varieties. The Liberty variety of blueberry used in this paper is a cross between Brigita and Eliot, which are both northern highbush blueberry varieties with strong cold tolerance, large sweet fruit, and long ripening periods, etc. The Liberty variety of blueberry inherits their excellent characteristics, and at the same time has the advantages of a tall tree, uniform fruit, and strong resistance to pests and diseases. The Liberty variety of blueberries inherits their excellent characteristics. Therefore, the Liberty variety is a representative variety (or population) because it is one of the most widely cultivated northern highbush blueberry varieties in the world, and it can reflect the general characteristics and patterns of the northern highbush blueberry population. So to some extent it can be assumed that the conclusions of this paper apply to the vast majority of northern highbush blueberries. In response to your question, we also provide a future outlook in the conclusion: In future research, we will consider testing the proposed shelf-life prediction model on multiple varieties or genotypes of blueberries to validate its usefulness in evaluating other varieties or genotypes and compare its prediction results with those of other models or methods in order to increase the generalisability of the results and dissemination of the research.

Q2: How long was the storage time of the blueberries and what were the storage conditions from the moment of harvest to the start of the research (T0)? Under what conditions were the berries transported to the laboratory?

A2: We thank the reviewers for their valuable comments and suggestions on our study. We agree with your views and have added a description in section 2.1. The storage time and conditions of blueberries were as follows: fresh "Liberty" blueberries were picked from a blueberry plantation in Dandong, Liaoning Province, and then immediately wrapped in ice packs and shipped in insulated containers to maintain the freshness and quality of the blueberries. Since our lab is in Harbin, which is far away from the picking site, we chose the fastest way of transport, i.e. by air. We loaded the insulated boxes onto the plane and kept them at a refrigerated temperature (0°C~4°C) on the plane. After shipping to the laboratory, the blueberries were pre-cooled in a refrigerated room at 10°C~12°C for 10h-12h to reduce the respiration rate and water loss of the blueberries. In summary, the storage time of blueberries from picking to the beginning of the study (T0) was about 14h-16h, which included 2h-4h of airfreight time and 10h-12h of pre-cooling time. The storage conditions were a refrigerated temperature of 0°C-4°C and a relative humidity of 90% ~95%.

Q3: The number of people who choose to conduct sensory assessment is too small to be convincing. What ages, sex and segments were the panelists?

A3: We thank the reviewers for their valuable comments and suggestions on our study. We take the reviewers' questions about the sensory evaluation section very seriously. We added the following to the sensory evaluation section:
(1) The number and characteristics of individuals selected for sensory evaluation were as follows: we selected 10 trained laboratory team members as evaluators, all of whom were blueberry enthusiasts and consumers and had the ability and experience to identify the quality and taste of blueberries. The age, gender and class of the evaluators are shown in Table 1, and they are all from different professions, academic qualifications and occupations, with a certain degree of representativeness and diversity.
(2) The methods and steps of sensory evaluation were as follows: a. We used a 9-point scale, with 1 being extremely disliked, 5 being average, and 9 being extremely liked. Evaluators cleaned their mouths before scoring, tasted each sample in order, and rinsed their mouths with warm water after tasting; b. We evaluated each sample in terms of the following three sensory aspects: appearance, flavour, and taste. Appearance included outer skin gloss, weightlessness, and degree of spoilage; flavour included aroma, sweetness, and acidity; and texture included crunchiness, chewiness, and firmness. c. We conducted sensory evaluations prior to the measurement of physicochemical indexes in order to avoid the influence of physicochemical indexes measurement on sensory evaluations. We carry out sensory evaluation once a day at the same time in the morning in a fixed order, and each sample is scored by 10 evaluators, and then the average score is calculated.
Regarding your question that the number of organoleptic evaluators is too small to be convincing, we believe that. We believe that:
(1) Referring to literature [8] and literature [11], it can be concluded that 10 trained evaluators are sufficient to reflect the sensory quality of blueberries and consumer preferences;
(2) The evaluators selected for this paper are trained laboratory panelists who are blueberry enthusiasts and consumers and have the ability and experience to identify the quality and taste of blueberries, as shown in Table 1, and they are representative and diverse;
(3) In this paper, we used the 9-point scale method, which is a commonly used sensory evaluation method that can effectively distinguish sensory differences between different samples and has high reliability and repeatability (Ref. [30]);
(4) As can be seen from Fig. 12, the sensory evaluation scores in this paper are strongly correlated with the results of physicochemical index measurements (SSC, total colour difference, PH, weight loss) and TPA test results (hardness), thus indicating that the sensory evaluation results in this paper have a certain degree of objectivity and validity.
In summary, I believe that 10 trained evaluators are sufficient to solve the problem of sensory evaluation and to obtain convincing results. I hope the reviewers will understand and accept our point of view. Thanks again to the reviewers for their rigour and professionalism.

Q4: Figures 5,12,13,14. - must be improved, the graphs are too small and illegible.

A4: Thank you to the reviewers for their valuable comments and suggestions. We strongly agree with your views and have made changes accordingly.

Q5: Tables 8,9. - must be improved, there is a lack of data in the paper (?) may be you should give them in horizontal layout.

A5: Thank you to the reviewers for their valuable comments and suggestions. We strongly agree with your views and have made changes accordingly.

Q6:"This suggests that low temperature storage can delay the ageing process of blueberries by preserving anthocyanins, and that 0°C storage is more effective than 4°C."  this is obvious, your experiment does not bring anything new on this topic.

A6: Thank you to the reviewers for their valuable comments and suggestions. We strongly agree with your views and have made changes accordingly.

Q7: Discussion - there is the lack of propre discussion of the results, the discussion is not correct, in this chapter the authors should carry out a comparative analysis of their research results with the research of other authors, this chapter needs to be improved. Maybe you could discuss the obtained results to other papers discussing the quality of blueberries “Liberty” and the changes during storage. This could improve this section and improve the reliability of the proposed prediction model.

A7: We thank the reviewers for their valuable comments and suggestions on our study. We take the reviewers' questions about the discussion section very seriously. We have added the following to the Discussion section: 

Shelf-life is a complex trait that is greatly influenced by genotype and variety. This study used ‘Liberty’ blueberry as the research object, which was obtained by crossing ‘Brigita’ and ‘Eliot’. Both ‘Brigita’ and ‘Eliot’ are northern highbush blueberry varieties, which have characteristics such as cold resistance, large and sweet fruits, and long maturation period [31]. ‘Liberty’ blueberry inherited their excellent characteristics, and also had advantages such as tall tree body, uniform fruit size, and strong resistance to diseases and pests [32]. These characteristics make ‘Liberty’ blueberry suitable for long-term storage and transportation, and also provide a good basis for shelf-life prediction. However, the shelf-life prediction problem of ‘Liberty’ blueberry has not been well studied before. Therefore, this paper proposed a novel shelf-life prediction model based on MRMR feature selection and GDEDBO-BPNN algorithm, and achieved high prediction accuracy at three temperatures of 0, 4, and 25 °C. The proposed model outperformed other models or methods reported for different crops in previous studies.

The impact of different input indicators on the model's prediction accuracy may vary. Figure 12 shows the correlation analysis results of different indicators and shelf-life at 0, 4, and 25 °C. Weight loss and firmness had the highest correlation with shelf-life at these temperatures, respectively. Thus, these two indicators could predict the shelf-life of blueberries more accurately than others. This was in agreement with Huang et al., who built blueberry freshness prediction models based on gas information and machine learning algorithms (BP, RBF, SVM, and ELM) at different temperatures and reported that SVM achieved the highest prediction accuracy of 94.01% [8]. However, the proposed model in this paper had a higher prediction accuracy of 96.73%, which might be attributed to the genotype differences that affected the quality change patterns and shelf-life of blueberries, and consequently the model performance.

Owoyemi et al. applied four models (MLR, SVR, RF, and XGBoost) to predict the shelf-life of ‘Rustenburg’ navel oranges at different temperatures and found that XGBoost performed the best with RMSE and R² of 0.195 and 0.914, respectively, on the whole dataset [30]. Zhang et al. used two methods (PLS and ANN) to establish the shelf-life prediction models of postharvest apples at 4 °C and 20 °C and found that ANN outperformed PLS with the optimal RMSE and R² of 3.51 and 0.991, respectively, under multiple indicators [6]. However, the proposed model in this paper had a higher prediction accuracy Which achieved the optimal RMSE and R² of 0.037 and 0.999, respectively, on the test set. The possible reasons for this are as follows:

(1) Species differences. The quality change patterns and shelf-life of fruits are influenced by their species characteristics, such as moisture content, antioxidant capacity, cell wall structure, sugar-acid ratio, etc. Generally speaking, fruits with higher moisture content, stronger antioxidant capacity, more stable cell wall structure, and moderate sugar-acid ratio have better quality and longer shelf-life. According to previous studies, ‘Liberty’ blueberry is a high-quality variety with high moisture content (about 85%), strong antioxidant capacity (about 13.8 mmol/100 g), stable cell wall structure (about 0.5%), and moderate sugar-acid ratio (about 15.6) [32]. However, ‘Rustenburg’ navel orange is a low-quality variety with low moisture content (about 60%), weak antioxidant capacity (about 2.4 mmol/100 g), unstable cell wall structure (about 1.2%), and high sugar-acid ratio (about 25.4) [30].

(2) Improper feature selection. Owoyemi et al. did not perform feature selection and might have redundant features. Zhang et al. used PLS to select features, but this method only considered the linear correlation between features and responses, without considering the redundancy and nonlinearity among features. However, the MRMR algorithm used in this paper could simultaneously consider relevance, redundancy, and nonlinearity among features. Therefore, it obtained higher prediction accuracy.

(3) Limitations of the models themselves. Owoyemi et al.'s XGBoost model had problems such as sensitivity to noise and outliers, overfitting, tedious hyperparameter tuning, etc., which affected its generalization ability. Zhang et al. used PLS to build the quality change and shelf-life prediction models of apples, but PLS assumed a linear relationship between features and responses, while there might be a nonlinear relationship in reality. They also used ANN to optimize the parameters of PLS model, but ANN was prone to fall into local optima and was sensitive to parameter selection. However, the BPNN model used in this paper could handle nonlinear and high-dimensional data, and used GDEDBO algorithm to optimize its parameters to avoid falling into local optima and minimize prediction error. Therefore, it obtained more satisfactory prediction results.

Q8: lines 832-838 are not conclusions but statements, in this section the results shouldn't be given, lines 836-838 - this is known and obvious.

A8: Thank you to the reviewers for their valuable comments and suggestions. We strongly agree with your views and have made changes accordingly.

 

Round 2

Reviewer 1 Report

The authors took into account all suggestions proposed by the reviewers.

Reviewer 2 Report

Dear Authors,

thanks for all the answers and explanations, this version is acceptable for publication, in my opinion. In addition, my research concerns the quality of fresh and stored food and your work inspired me to similar research. Thank you and best regards.