Non-Destructive Estimation of Fruit Weight of Strawberry Using Machine Learning Models
, Na Eun Kim 3, Nibas Chandra Deb 3, Bolappa Gamage Kaushalya Madhavi 3 and Hyeon Tae Kim 3,*
Round 1
Reviewer 1 Report
In this paper, a machine learning model is used to predict the fruit weight of strawberries using the number of pixels in the images of three types of strawberries. Simple linear regression (LR) and nonlinear regression (i.e., support vector regression (SVR)) models. The prediction results of the two models were compared, and their accuracy was verified. The accuracy of LR model in the training and testing stages reached 96.3% and 89.6% respectively.. Finally, the expectation of increasing strawberry varieties as samples to increase the model performance was proposed.
The key innovations are:
1） constructed a prediction model of strawberry fruit weight by using machine learning model
2） Proposed a non-destructive, time-saving, cost-effective new method for regular monitoring of fruit weight 。
The paper is well presented. However, before further consideration of acceptance, several questions need be explained, and some revisions are necessary:
1）The title of the article is chaotic, the subtitle 3.3 is followed by the subtitle 3.6, and the chapter 4 is missing. It needs to be reviewed again by the author
2）The author emphasizes that the feature of the article is non-destructive prediction of weight and gives a model of picture pixels and strawberry weight, but the acquisition of picture pixels not only takes strawberries but also makes a lot of preparations and fixes the focal length. Is it necessary to get pixel pictures in this way to predict the weight in the future? Since the strawberries are picked, why can't they be weighed?
3）Lines 210-214 “After testing on a number of SVR structures applying the three kernel functions i.e., polynomial, sigmoid functions and radial basis function (RBF) with gamma (γ), penalty parameter of the error term (C) and epsilon (ε), in this study, as a kernel type, we decided to use a radial basis function (RBF) with γ = 0.5, C = 50 and ε = 0.1.”What are the results of the test and how to draw a conclusion that these parameters are used? Are these parameters universal in predicting strawberry weight?
4）Lines 293-295，“Lee et al. [33] established a linear regression model to predict strawberry volume with coefficient of determinations were 0.866 and 0.603 in training and testing stages, respectively”Is there any improvement in the model of this study? How to add the comparison content of these two models?
5）Lines 361-362，“Similar results were also reported by Agulheiro‐Santos et al. [67] and Guo et al. [68]. ”Is there any optimization and new progress in the conclusion of this study?
Author Response
Response to Reviewer 1 Comments
Comments and Suggestions for authors
In this paper, a machine learning model is used to predict the fruit weight of strawberries using the number of pixels in the images of three types of strawberries. Simple linear regression (LR) and nonlinear regression (i.e., support vector regression (SVR)) models. The prediction results of the two models were compared, and their accuracy was verified. The accuracy of LR model in the training and testing stages reached 96.3% and 89.6% respectively. Finally, the expectation of increasing strawberry varieties as samples to increase the model performance was proposed.
The key innovations are:
1） constructed a prediction model of strawberry fruit weight by using machine learning model
2） Proposed a non-destructive, time-saving, cost-effective new method for regular monitoring of fruit weight.
The paper is well presented. However, before further consideration of acceptance, several questions need be explained, and some revisions are necessary:
Response: Dear reviewer, thank you for considering our manuscript and for the precious time you spent reviewing it. We made the changes in the manuscript according to the suggestion and comments provided. Overall, we corrected and tried to clarify the mistakes identified throughout the manuscript.
Point 1: The title of the article is chaotic, the subtitle 3.3 is followed by the subtitle 3.6, and the chapter 4 is missing. It needs to be reviewed again by the author
Response 1: Thanks for your comment. We are sorry and sincerely apologize for not being able to address the comments regarding the title of manuscript. All the authors as well as other peer-reviewer considered the current title of the manuscript is well stated.
Regarding the subtitle and chapter number, we corrected it in our revised manuscript (mentioned on page number 12 and line 363 of the manuscript and page number 13 and line 402 of the manuscript)
Point 2: The author emphasizes that the feature of the article is non-destructive prediction of weight and gives a model of picture pixels and strawberry weight, but the acquisition of picture pixels not only takes strawberries but also makes a lot of preparations and fixes the focal length. Is it necessary to get pixel pictures in this way to predict the weight in the future? Since the strawberries are picked, why can't they be weighed?
Response 2: Thank you for your valuable comments. In the current study, we calculated the pixel numbers of strawberry using a light chamber of dimension (80 cm × 80 cm × 80 cm) and for this case, the distance between the camera lens and strawberry fruit is maintained at 80 cm. The developed model can be used in the strawberry cultivation field in maintaining the distance between the camera and fruit by 80 cm (mentioned on page number 4 and lines 151-152 of the manuscript). The result suggested that it is possible to predict fruit weight using pixel numbers. In further experiments, we planned to use a depth-sensing camera, so that the distance between the fruit and the camera can be verified.
Since the model development period is a part of the supervised learning process, therefore, we need to use the actual (measured) fruit weight of strawberries as a dependent variable. That’s why we need to pick up fruit from strawberry plants for only model development time. However, it is not required to pick up during the application in real fields.
Point 3: Lines 210-214 “After testing on a number of SVR structures applying the three kernel functions i.e., polynomial, sigmoid functions and radial basis function (RBF) with gamma (γ), penalty parameter of the error term (C) and epsilon (ε), in this study, as a kernel type, we decided to use a radial basis function (RBF) with γ = 0.5, C = 50 and ε = 0.1.”What are the results of the test and how to draw a conclusion that these parameters are used? Are these parameters universal in predicting strawberry weight?
Response 3: In this experiment, we decided to use a radial basis function (RBF) with γ = 0.5, C = 50 and ε = 0.1. These results were obtained when the model was hyper-tuned using GridSearchCV() function in the sklearn library in python. This function trains the model with various hyper-tuning parameters and returns the optimized parameters for the model that has the highest performance (highest R2 and lowest RMSE). According to the optimized result of the hyper-tuning process, we decided to use those parameters in the current study
For this experiment, these hyper-tuning parameters gave the best result, however, these parameters are not universal and may change according to the dataset provided which includes fruit types, cultivar, etc.
Point 4: L140: Lines 293-295“Lee et al. [33] established a linear regression model to predict strawberry volume with coefficient of determinations were 0.866 and 0.603 in training and testing stages, respectively”Is there any improvement in the model of this study? How to add the comparison content of these two models?
Response 4: In the study conducted by Lee et al. [40], they developed a linear-based regression model to predict the volume of strawberries and the prediction accuracy was 86.6% in training and 60.3% in testing. However, in our study, the main objective is to predict fruit weight using linear and nonlinear (SVR) models. Our linear model has an accuracy of 96.3% in training and 89.6% in the testing stage which is higher than Lee et al. [40] findings. However, we cannot directly compare the results of the two studies as the target/objective of both studies are different (former: volume and later: weight) (mentioned on page number 9 and lines 306-308 of the manuscript).
Point 5: Lines 361-362，“Similar results were also reported by Agulheiro-Santos et al. [67] and Guo et al. [68]. ”Is there any optimization and new progress in the conclusion of this study?
Response 5: In the study conducted by Agulheiro-Santos et al. [75] and Guo et al. [76], they tried to predict the soluble solid content of strawberries using an image processing technique. In their study, they concluded that the prediction accuracy for the strawberry cultivars with irregular shapes was lower compared to the strawberry with regular/uniform shapes. Similar results were also obtained in the current study, where the strawberry cultivars with irregular shapes have lower accuracy in prediction of fruit weight compared to the regular/uniform shape (mentioned on page number 12 and lines 374-375 of the manuscript).
Author Response File: 
Author Response.docx
Reviewer 2 Report
In the current manuscript, Basak et al. have developed both linear and nonlinear-based machine learning models to predict fruit weight and comparisons their performance. The main objectives of this present study are three-fold: first, to determine the biometrical characteristics i.e., length, diameter, and weight at the six ripening periods of three strawberry cultivars; second, to acquire pixel numbers of strawberry images using image processing techniques and finally, to develop LR and SVR-based models using the pixel numbers for estimating fruit weight of strawberries. Although the topic is attractive, there are some concerns that should be addressed.
-Generally, the manuscript is well organized but there are some typographical and grammatical errors.
-The paper title is well stated, it is informative and concise.
-Abstract is well structured.
-The introduction was not well written, and it is too briefly presenting the subject and research problem. For instance, lines 62-63, please provide short description of machine learning and some examples of machine learning in agricultural and plant science such as plant breeding (https://doi.org/10.1016/j.isci.2020.101890), in vitro culture (https://doi.org/10.1007/s00253-020-10978-1), stress phenotyping (https://doi.org/10.1016/j.tplants.2015.10.015), stress physiology (https://doi.org/10.1371/journal.pone.0240427), plant system biology (https://doi.org/10.1007/s00253-022-11963-6), plant identification (https://doi.org/10.1016/j.compag.2016.07.003), pathogen identification (https://doi.org/10.1094/MPMI-08-18-0221-FI).
-Material and research methods are presented appropriately. The experimental setup and the description in the methods section are well structured, and the statistical analysis is correctly performed.
-The results obtained in this study are interesting. Results presented correctly.
-In general, the discussion of the results and conclusions are correct, but not sufficient. Discussion should be improved. The authors should discuss the uncertainties in machine learning and how they tackle this problem.
Author Response
Response to Reviewer 2 Comments
Comments and Suggestions for authors
Generally, the manuscript is well organized but there are some typographical and grammatical errors.
Response: Dear reviewer, thank you for considering our manuscript and for the precious time you spent reviewing it. We made the changes in the manuscript according to the suggestion and comments provided. Overall, we corrected and tried to clarify the mistakes identified throughout the manuscript.
Point 1: The paper title is well stated, it is informative and concise.
Response 1: Thanks for your comment.
Point 2: Abstract is well structured.
Response 2: Thank you for your invaluable comment.
Point 3: The introduction was not well written, and it is too briefly presenting the subject and research problem. For instance, lines 62-63, please provide short description of machine learning and some examples of machine learning in agricultural and plant science such as plant breeding (https://doi.org/10.1016/j.isci.2020.101890), in vitro culture (https://doi.org/10.1007/s00253-020-10978-1), stress phenotyping (https://doi.org/10.1016/j.tplants.2015.10.015), stress physiology (https://doi.org/10.1371/journal.pone.0240427), plant system biology (https://doi.org/10.1007/s00253-022-11963-6), plant identification (https://doi.org/10.1016/j.compag.2016.07.003), pathogen identification (https://doi.org/10.1094/MPMI-08-18-0221-FI).
Response 3: Thanks for your suggestion. We checked and corrected it in our revised manuscript (mentioned on page number 2 and lines 62-76 of the manuscript).
Point 4: Material and research methods are presented appropriately. The experimental setup and the description in the methods section are well structured, and the statistical analysis is correctly performed
Response 4: Thanks for your comment.
Point 5: The results obtained in this study are interesting. Results presented correctly.
Response 5: Thank you for your invaluable comment.
Point 6: In general, the discussion of the results and conclusions are correct, but not sufficient. Discussion should be improved. The authors should discuss the uncertainties in machine learning and how they tackle this problem.
Response 6: Thanks for your suggestion. We checked and corrected it in our revised manuscript (mentioned on page number 6 and lines 193-197, and page number 12 and lines 390-395 of the manuscript).
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
The Authors have sufficiently reflected what the reviewer concerns. I have no further questions.
Author Response
Thank you for your comments
Reviewer 2 Report
All my comments have been addressed. I think that the current version of the manuscript can be published in Agronomy.
Author Response
Thank you for your comments
