Detection of Aspergillus flavus in Figs by Means of Hyperspectral Images and Deep Learning Algorithms
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe work presented in the Manuscript, entitled „ Detection of Aspergillus flavus in figs by means of hyperspectral images and deep learning algorithms „.There are shortcomings and modifications that should be included in order to enhance the final manuscript for the readers.
· The introduction of the abstract from line 1 to line 7 should be shorted.
· The important results should be added in abstract.
· The conclusion of the abstract should be written at the end of abstract.
· More citations should be added from line 20 to line 24.
· More citations should be added from line 28 to line 33.
· The introduction dealt with hyperspectral images and deep learning algorithms in a very superficial way. In fact, the writing in the introduction must focus on these parts in a deeper way.· What is the novelty of the work?. What is new in your work that makes a difference in the body of knowledge?.
· Please remove the sentence from line 87 to line 90.
· Background should be added in mixed with introduction.
· Please remove the sentence in line 204.
· Figure 3 must be improved.
· In figure 3, the noise of the curve of spectral bands should be corrected. Then the data should be analysed again.
· From line 216 to line 221 should be added in result section.
· Figure 7 and 9 should be improved.
· There is no discussion section to discuss the results in the manuscript.
· Please, write the practical applications of your work in a separate section, before the conclusions and provide your good perspectives.
· Please write about the limitations of this work in details in conclusion section.
Comments on the Quality of English LanguageMinor editing of English language required.
Author Response
Firstly, we appreciate all the reviewer’s helpful comments. Next, we would like to clarify the main point of this work. These are the results of a first research work within a wider research study, where we will go deeper into our main purpose, which is to provide AI tools that can identify in an early phase the infection produced by aflatoxins such as Aspergillus flavus in fresh figs. We address fresh figs because they are perishable and its consumption occurs in a short period of time, so early detection is crucial to prevent them from entering the food chain. With this aim in mind, we have used two different architectures of neural networks that have already produced good results in other research projects.
As a first step, our aim is to extend this study, go deeper and collect more hyperspectral images from different harvests. Fresh and dried figs have a major impact on the national and regional economy. However, while dried figs have been studied extensively, fresh figs have not received the same attention. We believe that more research is needed, given the emerging market.
All changes in the paper are highlighted in red.
Reviewer 1:
The work presented in the Manuscript, entitled „ Detection of Aspergillus flavus in figs by means of hyperspectral images and deep learning algorithms „.There are shortcomings and modifications that should be included in order to enhance the final manuscript for the readers.
Comment 1: The introduction of the abstract from line 1 to line 7 should be shorted.
Response 1: Following the reviewer’s comments, we have shortened this paragraph in lines 1-4.
Comment 2: The important results should be added in abstract.
Response 2: We appreciate the reviewer’s comments. We have included a more extended explanation about the most important results between lines 14-23.
Comment 3: The conclusion of the abstract should be written at the end of abstract.
Response 3: Thank you for your valuable appreciation, we have included a brief conclusion in the abstract between lines 20-23
Comment 4: More citations should be added from line 20 to line 24.
Response 4: Following the reviewer’s comment, we have included three different references to support the ideas mentioned about the economic cost and the risk for human health. These references are between lines 28-31.
Comment 5: More citations should be added from line 28 to line 33.
Response 5: Thank you for your suggestion. We have added a new representative reference that clearly describes the conditions required for the fungus to grow. This reference is in line 37.
Comment 6: The introduction dealt with hyperspectral images and deep learning algorithms in a very superficial way. In fact, the writing in the introduction must focus on these parts in a deeper way.
Response 6: Thank you for your valuable appreciation, we have included a more extended paragraph about the basics of hyperspectral images and deep learning algorithms to analyze hyperspectral images. This information is included between lines 77-100
Comment 7: What is the novelty of the work?. What is new in your work that makes a difference in the body of knowledge?.
Response 7: Thanks to the reviewer for pointing out this. We would like to explain why we consider this research work worth to be published.
The presence of aflatoxins has been extensively studied in nuts, cereals, dried fruits (e.g. dried figs), cotton seeds, coffee, tea, etc. as well as other products made from these raw materials. Infection of A. flavus and A. parasiticus in dried figs usually occurs during transport, storage and processing and has until recently been cultivated in arid areas with low humidity conditions. However, the need to increase production given the emerging market, in this case for both dried and fresh figs, has led to the introduction of irrigated cultivation and thus changing temperature and humidity conditions. This leads to an increased risk of aflatoxin infection in fresh products entering the food chain. In view of this new scenario, we consider this work to be a first step in the development of tools for early detection.
The growth cycle of aflatoxin is between 3 and 5 days. Even if figs are intended to be dried, introducing fruit infected with aflatoxin into the process can cause contamination of other fruits. Therefore, detecting aflatoxin when the product is fresh is considered crucial, not only for direct consumption but also for further processing.
This last paragraph has been included in lines 39-42.
Comment 8: Please remove the sentence from line 87 to line 90.
Response 8: We follow the reviewer’s suggestion and this paragraph has been removed
Comment 9: Background should be added in mixed with introduction.
Response 9: We appreciate the reviewer’s suggestion. However, due to the delimitation of the background section, especially the application to agriculture as the backbone, we would like to keep both sections.
Comment 10: Please remove the sentence in line 204.
Response 10: According to the reviewer’s suggestion, the sentence has been removed.
Comment 11: Figure 3 must be improved.
Answer: Following the reviewer’s suggestion, Figure 3 has been improved.
Comment 12: In figure 3, the noise of the curve of spectral bands should be corrected. Then the data should be analysed again.
Response 12: We appreciate the reviewer’s comments. We would like to explain that the images were pre-processed with black/white correction to remove noise. This black/white correction has been applied during the pre-processing phase and we have added more detailed information about it in lines 222-232.
Comment 13: From line 216 to line 221 should be added in result section.
Response 13: Thank you for the comment. We agree with the reviewer and we have moved this paragraph at the beginning of the “Experimental result” section in lines 349-354.
Comment 14: Figure 7 and 9 should be improved.
Response 14: We appreciate the reviewer’s suggestion. We have improved both figures. We expect them to meet the quality criterion.
Comment 15: There is no discussion section to discuss the results in the manuscript.
Response 15: We appreciate the reviewer’s comment. Maybe it was not clear, but the conclusion section included discussion and conclusion. Now, we have divided the conclusion section in two different section: “Discussion” and “Conclusion”.
Comment 16: Please, write the practical applications of your work in a separate section, before the conclusions and provide your good perspectives.
Response 16: Following the reviewer’s suggestion. Our response to the reviewer will be as follow:
To describe the practical applications of this study, it is worth mentioning that the growth cycle of aflatoxin is between 1 and 5 days. Even if figs are intended to be dried, introducing fruit contaminated with aflatoxins into the process can cause contamination. Therefore, detecting aflatoxin when the product is fresh is considered crucial, not only for direct consumption but also for further processing.
This research work can be considered as the beginning of a research line especially aimed at analyzing and designing AI tools to facilitate the early detection of aflatoxins in fresh figs.
We have included the “Practical applications” section in lines 448-462.
Comment 17: Please write about the limitations of this work in details in conclusion section.
Response 17: Thank you for the reviewer’s suggestion.
We have some limitations related to the amount of samples analyzed. However, new samples from the last harvest are being taken and will be included in the following research works. Additionally, a relevant number of samples have been frozen to have enough samples to make up a huge dataset to train the following models. The main drawback is that the current models cannot be used in a production environment due to the limited availability of specific hardware and the need for more in-depth training.
This information has been included in the conclusion section, lines 492-497.
Reviewer 2 Report
Comments and Suggestions for AuthorsThe paper entitled “Detection of Aspergillus flavus in figs by means of hyperspectral images and deep learning algorithms” proposed ML methods to detect aspergillus flavus in figs using hyperspectral data, and concluded that ML and hyperspectral data could help to reach higher classification accuracy with no destruction of samples. Topic is interesting and results could provide reference for those working in food quality monitoring. However, I have some concerns that need to addressed during revision. I list my comments/suggestions for further improvements as follows.
1. Generally, hyperspectral data contains a lot of redundant information, why didn’t you consider to extract the specific bands that can represent aspergillus flavus infection which can reduce the data volume?
2. Line 168, about the resolution of hyperspectral camera, how many pixels in the height and width?
3. Lines 174-176, what are the average level of UFC concentration of real aspergillus flavus infected figs? The designed concentration difference is only 2 UFC/ml, can this represent the actual infection situation?
4. Lines 184-186, could you give more information on the data pre-processing? Did you make the radiometric correction to get the spectral reflectance? I suggest to clarify this process in detail.
5. Line 187, Since the inoculation procedure was carried out by immersion, will there be a significant difference between the inoculated samples and the real infected figs?
6. Figure 2, cannot see the legend clearly.
7. Line 205, “is” -> “it is”
8. Figure 3, in the tile please clarify the hyperspectral feature of which fig pixel, normal or infected figs? The right panel of Figure 3 was too small to see the wavebands.
9. Figure 4, too much spectral curves in the figure to see the differences, so I suggest to show these curves in different groups in sub-panels. In addition, can Figure 4 tell us the differences in the distribution of the bands in the different classes that can determine which percentile values will be chosen as inputs of DL?
10. Section 3.3, authors need to justify why taken FCNN and CNN in this study, for there are many ML models that are good at classification and regression.
11. Line 245, 448 is the number of spectral bands, not height.
12. What if removing wavelet transform in CNN? Could it still reach a higher accuracy?
13. The confusion matrixes and ROC curves indicated good performances of the proposed DL methods, however, a discussion section was missing to discuss the factors that affect the classification accuracy, limitation of the proposed methods, future work, and so on.
Author Response
Firstly, we appreciate all the reviewer’s helpful comments. Next, we would like to clarify the main point of this work. These are the results of a first research work within a wider research study, where we will go deeper into our main purpose, which is to provide AI tools that can identify in an early phase the infection produced by aflatoxins such as Aspergillus flavus in fresh figs. We address fresh figs because they are perishable and its consumption occurs in a short period of time, so early detection is crucial to prevent them from entering the food chain. With this aim in mind, we have used two different architectures of neural networks that have already produced good results in other research projects.
As a first step, our aim is to extend this study, go deeper and collect more hyperspectral images from different harvests. Fresh and dried figs have a major impact on the national and regional economy. However, while dried figs have been studied extensively, fresh figs have not received the same attention. We believe that more research is needed, given the emerging market.
All changes in the paper are highlighted in red.
Reviewer 2:
The paper entitled “Detection of Aspergillus flavus in figs by means of hyperspectral images and deep learning algorithms” proposed ML methods to detect aspergillus flavus in figs using hyperspectral data, and concluded that ML and hyperspectral data could help to reach higher classification accuracy with no destruction of samples. Topic is interesting and results could provide reference for those working in food quality monitoring. However, I have some concerns that need to addressed during revision. I list my comments/suggestions for further improvements as follows.
Comment 1: Generally, hyperspectral data contains a lot of redundant information, why didn’t you consider to extract the specific bands that can represent aspergillus flavus infection which can reduce the data volume?
Response 1: We appreciate the reviewer’s question.
We would like to say that in this work, we have extracted all the bands from the hyperspectral image and we work at a pixel level. A pixel is taken from the image corresponding to an area of the fig, where there must be contamination and all the spectral bands are used. This significantly reduces the redundancy of information, as 448 layers are extracted for one pixel. If we look at the spectral signature of one of the pixels, we can see that there are no anomalies in the curve, which indicates that there is no significant redundant information.
Comment 2: Line 168, about the resolution of hyperspectral camera, how many pixels in the height and width?
Response 2: We thank the reviewer’s question.The resolution of the hyperspectral came is 800x1024 and we have added this data in lines 198-199.
Comment 3: Lines 174-176, what are the average level of UFC concentration of real aspergillus flavus infected figs? The designed concentration difference is only 2 UFC/ml, can this represent the actual infection situation?
Response 3: We are really grateful for the comment of the reviewer, because we have made a mistake about the levels of concentration. We did not write the exponent properly in latex. The correct UFC concentration levels are 103, 105,107.
The UFC concentrations can vary between 102 and 107 UFC according to the severity of the infection and the environmental conditions such as temperature and humidity.
In this case, two different sets of figs were harvested in two different weeks. Each week, figs were divided into four groups: healthy controls and three groups were inoculated with the three different concentrations. After 24 hours after inoculation, hyperspectral imaging is started for the next 5 days, which represents the growth cycle of the aflatoxin. This process was repeated the second week with the second set of figs harvested.
The concentrations applied during the inoculation process really differ by 102 UFC/ml, which represents significant differences in contamination between the classes used.
Comment 4: Lines 184-186, could you give more information on the data pre-processing? Did you make the radiometric correction to get the spectral reflectance? I suggest to clarify this process in detail.
Response 4: We are grateful for this comment and we agree with the reviewer.
We have included additional explanations as well as the equation applied. We would like to explain that the images were pre-processed with black/white correction to remove noise. This black/white correction has been applied during the pre-processing phase and we have added more detailed information about it in lines 222-232.
Comment 5: Line 187, Since the inoculation procedure was carried out by immersion, will there be a significant difference between the inoculated samples and the real infected figs?
Response 5: We appreciate the reviewer’s question. Our response to the reviewer will be as follows:
There were no visible differences between the contaminated and uncontaminated samples throughout the experiment. Two sets of samples were collected over a two-week period, with each set gathered weekly. After collection, the samples were inoculated, and 24 hours later, hyperspectral imaging was initiated. Following the initial imaging, the samples were stored at 25 ºC in refrigeration and imaged again the next day. This process was repeated for one week with each set of samples. Despite varying contamination levels, no visible changes were observed in any of the samples during the experiment.
We have written more detail explanations in lines 206-214.
Comment 6: Figure 2, cannot see the legend clearly.
Response 6: We appreciate the reviewer's comment. However, Figure 2 has no legend. It may be that this comment is in relation to Figure 3. This figure has been enhanced.
Comment 7: Line 205, “is” -> “it is”
Response 7: Thank you for the reviewer’s comment. This mistake has been fixed.
Comment 8: Figure 3, in the tile please clarify the hyperspectral feature of which fig pixel, normal or infected figs? The right panel of Figure 3 was too small to see the wavebands.
Response 8: We appreciate the reviewer's comment. This figure has been enhanced.
Comment 9: Figure 4, too much spectral curves in the figure to see the differences, so I suggest to show these curves in different groups in sub-panels. In addition, can Figure 4 tell us the differences in the distribution of the bands in the different classes that can determine which percentile values will be chosen as inputs of DL?
Response 9: Thanks to the reviewer for pointing out this.
We agree with the reviewer that Figure 4 is a bit confusing because there are too many lines. We have therefore decided to show the zoom of a range of wavelengths, in particular 526 nm - 623 nm as an example and additional explanation about this wavelength range has been included in lines 262-265.
Comment 10: Section 3.3, authors need to justify why taken FCNN and CNN in this study, for there are many ML models that are good at classification and regression.
Response 10: We appreciate the reviewer’s comment. We agree with the reviewer, there are a lot of different ML models for classification and regression. We are following a line of research with several previous research works in which we used CNN for the analysis of data from other types of crops. In this work, we are following this line of research in the computer vision area because we have obtained really good results. We present these works below:
Rodríguez, F.J., García, A., Pardo, P.J. et al. Study and classification of plum varieties using image analysis and deep learning techniques. Prog Artif Intell 7, 119–127 (2018). https://doi.org/10.1007/s13748-017-0137-1
Miragaia, R.; Chávez, F.; Díaz, J.; Vivas, A.; Prieto, M.H.; Moñino, M.J. Plum Ripeness Analysis in Real Environments Using Deep Learning with Convolutional Neural Networks. Agronomy 2021, 11, 2353. https://doi.org/10.3390/agronomy11112353
Chávez, A. Vivas, M. J. Moñino and F. Fernández, "METSK-HD-Angeleno: How to predict fruit quality using Multiobjective Evolutionary learning of TSK systems," 2019 IEEE Congress on Evolutionary Computation (CEC), Wellington, New Zealand, 2019, pp. 1251-1258, doi: 10.1109/CEC.2019.8790268. keywords: {Sugar;Software tools;Prediction algorithms;Solids;Image color analysis},
Comment 11: Line 245, 448 is the number of spectral bands, not height.
Response 11: Thank you for pointing out the mistake. We have fixed it
Comment 12: What if removing wavelet transform in CNN? Could it still reach a higher accuracy?
Response 12: We appreciate the reviewer’s question. We would like to answer that the wavelet transform is used precisely to transform the problem into a computer vision problem. This allows us to obtain images of the spectrum representing the spectral signature in use. In this work, we have not studied if It is possible to obtain better results without the transform. The reason why we have applied this technique in this work is because the components of the research group obtained good results in other research work with similar characteristics.
There are other possibilities that we are going to consider and explore as part of further research based on these results and the line of research that we are working in.
- Patterns Detection in Glucose Time Series by Domain Transformations and Deep Learning. Jorge Alvarado, J. Manuel Velasco, Francisco Chavez, Francisco Fernández-de-Vega, J. Ignacio Hidalgo, Combining wavelet transform with convolutional neural networks for hypoglycemia events prediction from CGM data, Chemometrics and Intelligent Laboratory Systems, Volume 243, 2023, 105017, ISSN 0169-7439, https://doi.org/10.1016/j.chemolab.2023.105017.
Comment 13: The confusion matrixes and ROC curves indicated good performances of the proposed DL methods, however, a discussion section was missing to discuss the factors that affect the classification accuracy, limitation of the proposed methods, future work, and so on.
Response 13: We appreciate the reviewer’s comment, maybe it was not clear, but the conclusion section included discussion and conclusion. Now, we have divided the conclusion section in “Discussion” and “Conclusion”.
Reviewer 3 Report
Comments and Suggestions for AuthorsThank you for selecting me as a reviewer for this important research; it is an honor to contribute to the scientific community.
Comments for author File: Comments.pdf
nil
Author Response
Response to the reviewer’s comments:
Firstly, we appreciate all the reviewer’s helpful comments. Next, we would like to clarify the main point of this work. These are the results of a first research work within a wider research study, where we will go deeper into our main purpose, which is to provide AI tools that can identify in an early phase the infection produced by aflatoxins such as Aspergillus flavus in fresh figs. We address fresh figs because they are perishable and its consumption occurs in a short period of time, so early detection is crucial to prevent them from entering the food chain. With this aim in mind, we have used two different architectures of neural networks that have already produced good results in other research projects.
As a first step, our aim is to extend this study, go deeper and collect more hyperspectral images from different harvests. Fresh and dried figs have a major impact on the national and regional economy. However, while dried figs have been studied extensively, fresh figs have not received the same attention. We believe that more research is needed, given the emerging market.
All changes in the paper are highlighted in red.
Reviewer 3:
Comment 1: The introduction mentions the significance of aflatoxin detection, a more thorough explanation of the particular difficulties that fresh figs—as opposed to dried figs—present would be beneficial.
Response 1: Thanks to the reviewer for pointing out this. We would like to explain why we consider that analyzing fresh figs would be addressed:
Firstly, the presence of aflatoxins has been extensively studied in nuts, cereals, dried fruits (e.g. dried figs), cotton seeds, coffee, tea, etc. as well as other products made from these raw materials. Infection of A. flavus and A. parasiticus in dried figs usually occurs during transport, storage and processing. Figs are known to be historically cultivated in arid areas with low humidity conditions. However, the need to increase production given the emerging market, in this case for both dried and fresh figs, has led to the introduction of irrigated cultivation, and thus changing temperature and humidity conditions. This increases the risk of aflatoxin infection in fresh products entering the food chain. In view of this new scenario, we consider this work to be a step in the development of tools for early detection.
Secondly, the growth cycle of aflatoxin is between 3 and 5 days. Even if figs are intended to be dried, introducing figs infected with aflatoxin into the process can cause contamination of other fruits. Therefore, detecting aflatoxin when the product is fresh is considered crucial, not only for direct consumption but also for further processing.
Some additional information has been added in lines 39-42.
Comment 2: Describe how your strategy particularly tackles these issues and why it is better than the current approaches.
Response 2: Thank you for giving us the opportunity to explain our approaches.
This research work tries to put the first step addressing the early detection of A. Flavus infection in fresh figs. Our research work got good result using these methodologies on other fruits and we think this is a good starting point.
Currently, we are not sure if it is better than other approaches, but we have set a work line, where there is much research work to do. Actually, we are working on taking more samples from different harvests to validate our results. We are also designing the following steps by selecting the application of a set of different ML algorithms, analyzing different methodologies to reduce the dimensionality, improve sustainability and carry out a comparative analysis using other non invasive techniques.
Comment 3: The case for HSI and DL as the most practical approach may be strengthened by the thorough comparisons with other non-invasive aflatoxin detection techniques, such as UV imaging or other optical approaches.
Response 3: We appreciate the reviewer’s suggestion. Our respond to the reviewer will be as follows:
This is the first research work of a line of research, there is much research work to do. We consider further research to carry out these comparisons.
Comment 4: The hyperspectral picture acquisition process is briefly explained, but not enough information is provided about the surroundings at the time of the fig inoculation.
Response 4: We appreciate the reviewer’s question. Our respond to the reviewer will be as follows:
After collection, the samples were inoculated by immersion. This process has been carried out by immersing the area for about 3 seconds, according to the protocol in Cicytex, the organization that is in charge of this particular phase of the experiment. 24 hours later, hyperspectral imaging was initiated. Following the initial imaging, the samples were stored at 25 ºC in refrigeration and images were taken every 24 hours for five days. This process was repeated for one week with each set of samples. Despite varying contamination levels, no visible changes were observed in any of the samples during the experiment.
We have included more detail information in lines 206-215.
Comment 5: The more details on these variables would make it easier for them to comprehend how consistent and trustworthy the data is.
Response 5: Thank you for the reviewer’s comment. We have included more information about the process.
Comment 6: The process of standardizing the fungal inoculation across the various concentration levels requires more explanation. This may have an impact on how applicable the findings are to actual contamination situations.
Response 6: We are really grateful for the comment of the reviewer, because we have made a mistake about the levels of concentration. We did not write the exponent properly in latex. The correct UFC concentration levels are 103, 105,107.
The UFC concentrations can vary between 102 and 107 UFC according to the severity of the infection and the environmental conditions such as temperature and humidity.
In this case, two different sets of figs were harvested in two different weeks. Each week, figs were divided into four groups: healthy controls and three groups were inoculated with the three different concentrations. After 24 hours after inoculation, hyperspectral imaging is started for the next 5 days, which represents the growth cycle of the aflatoxin. This process was repeated the second week with the second set of figs harvested.
The concentrations applied during the inoculation process really differ by 102 UFC/ml, which represents significant differences in contamination between the classes used.
This protocol is applied by the Scientific and Technological Research Center of Extremadura (Cicytex).
Comment 7: While the confusion matrices are given effectively, a deeper review of the material would be beneficial.
Response 7: We appreciate the reviewer’s question. An explanation of the values of the confusion matrix, from our point of view from an engineering background, is given in a paragraph in the “Experimental results” section for both FCNN and CNN approaches in lines 365-373 and 416-421, respectively.
Comment 8: Strengthening the discussion section will need offering theories and perspectives on model behavior.
Response 8: Thanks for the reviewer’s suggestion. Our response to the review will be as follows:
We have followed the reviewers' suggestions. We have added more explanations in the sections "Experimental results" and "Conclusion". In addition, we have added two new sections - "Practical Applications" and "Discussion". We hope that all the information provided in these sections helps to strengthen the research we wish to carry out.
Comment 9: A brief mention is made of the ROC curves' AUC values. But more clarification on these qualities and their applications would be beneficial, especially when it comes to differentiating between tainted and healthy figs.
Response 9: We have followed the reviewer’s suggestion.
We included some comments on ROC and AUC results in lines 389-391 and 434-437 for the FCNN and CNN models, respectively.
Comment 10: An essential conversation about the environmental cost of AI models is introduced in this study.
Response 10: We appreciate the reviewer’s comment and we agree with the reviewer. We will continue addressing this important issue in further research.
Comment 11: There is not enough context provided for the outcomes. How do the energy and CO2 emissions compare to other AI applications or similar studies? Benchmarks would lend greater weight to the sustainability conversation.
Response 11: We appreciate the reviewer’s question. Our respond to the reviewer will be as follows:
We are addressing this issue in current research, so we will have multiple measurements about CO2 emissions and energy consumptions using different AI techniques on our own datasets. We think this will be a fear study.
Comment 12: In subsequent studies, it would be helpful to offer methods for lessening the environmental impact of these models.
Response 12: We appreciate the reviewer’s suggestion.
We agree with the reviewer and our research group is working on it from different points of view.
Comment 13: Future research directions might be better addressed in the conclusions.
Response 13: Thank you for this suggestion.
We have divided the conclusion section into “Discussion” and “Conclusion” and we have included some insights about our future research works. We have also written a “Practical applications” section.
Comment 14: More practical relevance would come from incorporating this technology into actual agricultural monitoring systems or food safety inspection procedures.
Response 14: Thanks for the annotation, we have included the following text in “Practical applications” section:
The incorporation of such tools into agricultural monitoring systems should become a reality and can be easily adapted to the Extremadura region. Currently, Extremadura exports 71,000 tonnes of figs worldwide, and the ability to detect infected figs in the production chain would enable the delivery of a higher-quality and healthier product. At present, these anomalies are identified at advanced stages of the production chain, resulting in the disposal of significant quantities of produce due to minimal contamination. Detecting contaminated figs at earlier stages of the production process would significantly improve product quality. (https://higosdealmoharin.com/tag/exportacion/)
This can give the reviewer some insights about the practical applications.
Comment 15: The grammatical mistakes and strange phrasings need to be checked.
Response 15: Thank you for pointing out the mistakes. We have made corrections and checked English trying to enhance it.
Comment 16: The pictures are useful, yet some—particularly the technical schematics of the neural network architectures—might be improved for easier reading.
Response 16: Thanks to the reviewer for pointing out this issue. We would like to say that the images have been improved.
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsBackground should be mixed with introduction.
The authors should combined the results and discussion in one section. Also the authors should compare thier Data with pervious studies. I see only results section without discussion with pervious studies.
Author Response
Firstly, we appreciate all the reviewer’s helpful comments during this review process.
All new changes in the paper are highlighted in red.
Comment 1. Background should be mixed with introduction.
Response 1. Following the reviewer’s suggestion, we have merged “Introduction” and “Background” sections. Some paragraphs have been a little reduced and the end of the merged section has been updated to include the summary paragraph from the previous Introduction section.
Only parts updated have been highlighted in red.
Comment 2. The authors should combined the results and discussion in one section. Also the authors should compare thier Data with pervious studies. I see only results section without discussion with pervious studies.
Response 2: We appreciate the reviewer’s comments. Our response to the reviewer will be as follows:
Concerning the discussion and conclusion sections: We have to point out that in the previous round, some reviewers suggested that we should split the Conclusion section into 2 separate sections: "Discussion" and "Conclusion". We accepted this suggestion and that is why there are two different sections.
Regarding the comparison of previous studies, it should be noted that there are several studies on A. flavus infection in dried figs, but we did not find any studies on fresh figs. On the other hand, we need to consider that the chemical composition of dried and fresh figs is completely different.
Additionally, in the agriculture sector, the comparison with the same experiment in different areas or latitudes is not fair because the climate conditions and soil composition are very different. In this preliminary work, we try to develop a set of experiments, using data collected from the same orchards. Further research using data from the same orchards will allow us to carry out a fair comparison.
This experiment can be tested using data collected from other areas with different climate conditions and soil composition to determine the goodness of the model.
We have added a paragraph in the conclusion section to clarify this point
Reviewer 2 Report
Comments and Suggestions for AuthorsAbout the comment 6, what does the colors in Figure 2 mean? It seems that the upper right corner is showing the information but too small to see.
The manuscript was improved significantly. I think it will be ready for publication after a minor edit about Figure 2.
Author Response
Firstly, we appreciate all the reviewer’s helpful comments during this review process.
All changes in the paper are highlighted in red.
Comment 1. About the comment 6, what does the colors in Figure 2 mean? It seems that the upper right corner is showing the information but too small to see. The manuscript was improved significantly. I think it will be ready for publication after a minor edit about Figure 2.
Response 1. We really appreciate the reviewer’s comment. This figure illustrates the use of the annotation tool 'LabelMe,' which was used to select the ROI for each fig analyzed. During segmentation, the tool assigns random colors to improve ROI visibility, but these colors carry no meaningful information.
To avoid any potential misunderstanding, we have changed the figure and removed the upper part.
Round 3
Reviewer 1 Report
Comments and Suggestions for Authors1- Practical applications should be added after discussion section.
2- Discussion section is very weak to accept.
Comments on the Quality of English LanguageMinor editing of English language required
Author Response
We appreciate the effort of the reviewers in this process. We are very grateful for the opportunity to improve the presentation of our research.
Comment 1: Practical applications should be added after discussion section.
Response 1: We have moved the "Practical Applications" section after the "Discussion" section, as suggested by the reviewer.
Comment 2: Discussion section is very weak to accept.
Response 2: We appreciate the reviewer's comment and have modified the Discussion section to include a deeper critical aspects that we believe needs to be addressed in future research. We hope that this section will meet the quality criteria you have requested.
Comment 3: Minor editing of English language required
Response 3: Thank you for the reviewer's suggestion. We have checked the English language and corrected any typos we found.