Development of Fluorescence Imaging Technique to Detect Fresh-Cut Food Organic Residue on Processing Equipment Surface

Round 1

Reviewer 1 Report

Introduction

#41  produce- I assume the correct word is product

#45-#46 Missing clarity of the sentence

#47 Repetition of the word Bacteria

#46-#47 Missing clarity of the sentence

#50 Paragraph starting at this line contains a lot of statements without citing any references.

#72 It doesn’t make much meaning of including the rest of the paragraph (starting from- Environmental applications…..) as the topic mentioned has no relation and this is written after the hyperspectral imaging for this project is already introduced before.          

#92 preperation- Correct the spelling

2.2. System and data acquisition

What is the imaging geometry used?

How did the reflections from the SSS is managed?

Need a detailed explanation of the hyperspectral fluorescence image acquisition process.

What thresholding is been used?

#226, #229, etc. What are these wavelength bands? For example, 1983, 827, 545 and 362? Isn´t this the thresholding value?

With details included in Table 1-6, is it necessary to have the same details in text explained in different sections? Major interpretations could be included.

Figure 6: How is the color image obtained? Is this a true color or just 3 band images?

Did the authors make a similar conclusion regarding the prominent wavelength in a previous study (Ref 6), however, this is not clearly mentioned in this manuscript?

Conclusion could be shortened, there is no need to explain every result but rather include only the major findings.

What effort is needed to extend this study to other agricultural products?

Is the data and code of this study available?

Author Response

We appreciate your comments. The response to the review comments is as follows.

1. #41  produce- I assume the correct word is product

Response: The word of ‘produce’ is selected to mean product that have been produced or grown, especially by farming.

2. #45-#46 Missing clarity of the sentence

Response: The sentence has been corrected as follows; Contamination on the surfaces of food processing equipment is one of the potential causes of pathogen transmission to finished products which is produced using this food-processing procedure.

3. #47 Repetition of the word Bacteria

Response: It has been corrected.

4. #46-#47 Missing clarity of the sentence

Response: The sentence has been corrected as follows; Food residues remaining on equipment surfaces can shield bacteria from sterilization and drying stress.

5. #50 Paragraph starting at this line contains a lot of statements without citing any references.

Response: It has been corrected.

6. #72 It doesn’t make much meaning of including the rest of the paragraph (starting from- Environmental applications…..) as the topic mentioned has no relation and this is written after the hyperspectral imaging for this project is already introduced before.          

Response: The paragraph (line 72-77) has been deleted.

7. #92 preperation- Correct the spelling

Response: It has been corrected.

8. What is the imaging geometry used?

Response: Band Interleaved by line (BIL) scanning method is used to acquire an image composed of 502 X 501 pixels (spatial X spatial) X 70 (spectral).

9. How did the reflections from the SSS is managed?

Response: There was no effect on the reflection because the fluorescence of the SSS instead of the reflectance of the SSS was used.

10. Need a detailed explanation of the hyperspectral fluorescence image acquisition process.

Response: The paragraph (line 112-120) has been corrected as follows; Hyperspectral line-scan images of the juice residues on SSS were obtained by moving the samples by 740 steps at 0.5 nm per step for an exposure time of 0.2 s in a darkened room. Each line-scan contained 502 x 501 spatial pixels obtained by pairwise binning (averaging) of the 1004 x 1002 available pixels in the spatial dimension, and 70 wavebands spanning 450–730 nm in the spectral dimension that were obtained by binning at approximately 4 nm intervals. Since the wavelength region after 730 nm is a region that is repaeated as the effect of second order, this region was removed by a filter (Kodak Wratten Gelatin Filter, No. 8). Through this acquisition process, hyperspectral image cubes composed of 502 x 501 x 70 (spatial x spatial x spectral) were collected using the in-house developed software [12, 38].

11. What thresholding is been used?

Response: Thresholding is used to discriminate between clean condition and residue using the relative fluorescence intensity (RFI) of the selected waveband.

12. #226, #229, etc. What are these wavelength bands? For example, 1983, 827, 545 and 362? Isn´t this the thresholding value?

Response: Yes, they (1983, 827, 545 and 362) are thresholding values. The wavelength bands have been explained at line 224 ‘close to 480nm’.

13. With details included in Table 1-6, is it necessary to have the same details in text explained in different sections? Major interpretations could be included.

Response: Yes, since they represent the analysis results applied with different algorithms, each should be explained. Major interpretations were included in line 312-321 for table1-2, in line 400-403 for table3-4, in line 436-438 for table5-6,

14. Figure 6: How is the color image obtained? Is this a true color or just 3 band images?

Response: The image is true color image obtained by a digital camera.

15. Did the authors make a similar conclusion regarding the prominent wavelength in a previous study (Ref 6), however, this is not clearly mentioned in this manuscript?

Response: The reference has been added in line 491.

16. Conclusion could be shortened, there is no need to explain every result but rather include only the major findings.

Response: Conclusions are revised. Line 465-470 (For both surfaces ~ at low concentrations.) and line 475-476 (The results ~ TWRA.) have been deleted.

17. What effort is needed to extend this study to other agricultural products?

Response: After investigating the fluorescence peaks of other agricultural products, the results of this study can be applied.

18. Is the data and code of this study available?

Response: It is available as a database for other agricultural products.

Author Response File: Author Response.docx

Reviewer 2 Report

The article presents research of great interest to the fresh cut fruit industry. Hyperspectral fluorescence imaging is a promising technique that can be implemented to complement the performance of currently practiced methods in the industry to ensure the safety of these products. The experimental design is adequate. However, for the manuscript to be published it is necessary to make some improvements:

- The introduction should be expanded by delving into the bases of fluorescence: excitation wavelengths and emission wavelengths of fruits and their juices reported in the bibliography.

- Line 47: delete 'Bacteria'

- Line 104: correct 'imaing device'

- Line 112: Why are you selecting excitation light at 365-nm? Explain in the text.

- Line 117: please, explain the meaning of 'removal of the influence of the second order'

- Line 122: in spite you refer to reference [43], explain briefly what the flat-field image is

- Line 127: please, give more details about the procedure for extracting spectra of the ROI: how many spectra of each sample? selecting a region from the centre to the limits of the droplet? selecting single points?

-Line 133: An algorithm? What algorithm? Please, explain.

- Figures 3, 4, 5: Average spectra of these figures are raw spectra or spectra extracted of I "the corrected relative hyperspectral fluorescence image" (according to equation 1). Explain. You are presenting only the average spectra, but you are using images, so it would be interesting to present all the spectra extracted from the ROI, which would allow us to assess the variability in the ROI. Please include the ROI spectra of at least one sample.

- ANOVA: it is necessary to explain in more detail

- Are you performing an ANOVA comparing each diluted sample (for each fruit) with the clean SSS (for each SSS)?

- Include in the Tables the F-values

- What is TH? threshold values for segregating diluted samples? Refers TH to intensity (in raw spectra) in the selected waveband?

- Show graphicaly the results of ANOVA, for example, with a box and whiskers plot, to indicate variability (upper and lower quartile, outliers).

- Tables 1 and 2: what does "No. of sample" mean?

- Tables 1 and 2: what does "No. of sample" mean?

- Table 7: Are you referring to binary images? How is precision calculated when images are evaluated? Is the value of each pixel in the ROI considered?

Author Response

We appreciate your comments. The response to the review comments is as follows.

1. The introduction should be expanded by delving into the bases of fluorescence: excitation wavelengths and emission wavelengths of fruits and their juices reported in the bibliography.

Response: The line 74-79 are revised as follows;

Hyperspectral fluorescence imaging is a sensitive optical technique that uses selected light excitation of a sample to induce light emissions from the sample at wavelengths different from the excitation light of 365 nm [6, 32,39]. In studies targeting detection of bacterial biofilm, feces, organic residues, and insects in agricultural produce for food safety, non-destructive and non-contact fluorescence imaging using fluorescence regions from 480 to 560 nm and from 670 to 690 nm has been demonstrated to be an effective technique for rapid detection of contaminants [6, 35-41].

2. Line 47: delete 'Bacteria'

Response: It has been corrected.

3. Line 104: correct 'imaing device'

Response: It has been corrected.

4. Line 112: Why are you selecting excitation light at 365-nm? Explain in the text.

Response: The sentence has been added in line 102-104 as follows;

To identify food components such as chlorophyll a, UV-A (365 nm) excitation light, whose fluorescence emission range is in the blue-to-near-infraredregion (up to 730 nm), was used [54].

5. Line 117: please, explain the meaning of 'removal of the influence of the second order'

Response: The sentence has been corrected as follows; Since the wavelength region after 730 nm is a region that is repaeated as the effect of second order, this region was removed by a filter (Kodak Wratten Gelatin Filter, No. 8).

6. Line 122: in spite you refer to reference [43], explain briefly what the flat-field image is

Response: It has been added in line 125 as follows;

a flat-field image shows that fluorescence was exhibited uniformly [43].

7. Line 127: please, give more details about the procedure for extracting spectra of the ROI: how many spectra of each sample? selecting a region from the center to the limits of the droplet? selecting single points?

Response: The sentence has been added in line 130-131 as follows;

The region of interest (ROI) consists of the region from the center to the boundary of the residue, and 15–25 pixels per droplet were selected in this region.

8. Line 133: An algorithm? What algorithm? Please, explain.

Response: The word has been corrected as follows; global algorithm. And the description has been added in line 143-146 as follows; Fruit residues were identified, and accuracies of detection were calculated by the algorithms based on single waveband and two-waveband ratio.

9. Figures 3, 4, 5: Average spectra of these figures are raw spectra or spectra extracted of I "the corrected relative hyperspectral fluorescence image" (according to equation 1). Explain. You are presenting only the average spectra, but you are using images, so it would be interesting to present all the spectra extracted from the ROI, which would allow us to assess the variability in the ROI. Please include the ROI spectra of at least one sample.

Response: A Figure representing the spectra extracted from the ROI area of the orange residue was added.

10. ANOVA: it is necessary to explain in more detail

Response: The description was added in line 134-138 as follows;

The spectra extracted from the ROI was divided into two groups, clean surface and residue, labelled as 0 and 1, respectively. One-way ANOVA was performed on all 70 wavebands in the 450–730 nm region to identify the difference between the two groups. The F-value obtained by ANOVA analysis for a single waveband and a ratio of two wavebands was used to select the best wavebands [32,40].

11. Are you performing an ANOVA comparing each diluted sample (for each fruit) with the clean SSS (for each SSS)?

Response: Yes, each sample of the residue was compared with the ROI area of clean SSS.

12. Include in the Tables the F-values

Response: The F-values has been added.

13. What is TH? threshold values for segregating diluted samples? Refers TH to intensity (in raw spectra) in the selected waveband?

Response: Yes, TH is the fluorescence intensity for the selected waveband and represents the threshold value for segregating residue from the stainless steel sheet.

14. Show graphicaly the results of ANOVA, for example, with a box and whiskers plot, to indicate variability (upper and lower quartile, outliers).

Response: ANOVA analysis results to distinguish between contaminated and clean regions are used to determine the F-value for each spectral wavelength and to select the wavelength with the maximum F-value. So it would not be appropriate to show a box and whiskers plot.

15 Tables 1 and 2: what does "No. of sample" mean?

Response: It was corrected as follows; Number of spectra.

16. Table 7: Are you referring to binary images? How is precision calculated when images are evaluated? Is the value of each pixel in the ROI considered?

Response: Yes, they related to binary images. But the values were considered the number of detected droplets, not pixels. If there is even one ‘1’ in the droplet, we counted that the residue was detected.

Author Response File: Author Response.pdf

Reviewer 3 Report

The paper ‘Development of fluorescence imaging technique to detect fresh-cut food organic residue on processing equipment surface’ is an attempt to apply monitoring technique using hyperspectral fluorescence images, to detect fruit residues on food-processing equipment surfaces. Authors provide a wide and interesting discussion regarding results obtained during experiments, but the paper needs some additional corrections to fulfill standards of ‘Applied sciences’ journal.

Minor comments:

1. There are several editorial and grammar errors (e.g. l. 25 ‘were are’, l. 46-47 ‘ Bacteria Bacteria’. Please recheck manuscript carefully to remove all of them
2. Figures 3, 4, 5 should be adjusted to monochrome presentation. If one print the paper, graphs are almost unreadable – one can’t recognize the description of the lines.
3. The quality of Figures are very poor – The quality should be higher or the way of results presentation should be changed

Major comments

1. The reader would have an impression that there is only one method to detect organic residues. The comparison of efficiency (the same data, results comparison and discussion) between method proposed by authors and other known methods e.g. described in introduction should be introduced.
2. The structure of the paper is unbalanced. The material and methods section should be extended. Flowchart of methodology for the development of the algorithm looks interesting, but authors should add the section regarding image recognition technique as well as detailed description of algorithms (complexity, diagram representation).
3. Potential reader is not familiar with machine learning terminology. Authors should add the information regarding description of dataset, validation technique, accuracy measurement strategy and definition. E.g. were the same droplets used in training and testing stage? How final rules were identified. Did you use any of machine learning techniques (you mentioned about accuracy, validation, etc.)
4. The experiment should be described in the way to allow the reader to repeat and validate results – the description of all parameters of the experiment, and each step should be included in the paper (e.g. MATLAB parameters, etc)
5. The discussion can confuse the reader. What is the reason to provide long discussion regarding residue detection algorithms for different fruits? Are there serious difference between these algorithms? If so, please describe that in details. This will be more valuable for the reader than description of the figures. Please cut such description or improve abstract and title by identification of different types of fruits.
6. Authors mentioned, that on the figures ‘the average spectra’ for each fruits are presented, but there are no statistical analysis. Can you provide such analysis on the figures to justify your conclusions?

Author Response

We appreciate your comments. The response to the review comments is as follows.

Minor comments:

1. There are several editorial and grammar errors (e.g. l. 25 ‘were are’, l. 46-47 ‘ Bacteria Bacteria’. Please recheck manuscript carefully to remove all of them

Response: It has been corrected.

2. Figures 3, 4, 5 should be adjusted to monochrome presentation. If one print the paper, graphs are almost unreadable – one can’t recognize the description of the lines.

Response: It has been corrected.

3. The quality of Figures are very poor – The quality should be higher or the way of results presentation should be changed.

Response: It has been corrected.

Major comments

4. The reader would have an impression that there is only one method to detect organic residues. The comparison of efficiency (the same data, results comparison and discussion) between method proposed by authors and other known methods e.g. described in introduction should be introduced.

Response: In line 50, we have already introduced the existing organic residues inspection method, and this study is a research that proposes a method to measure faster than the conventional method.

2. The structure of the paper is unbalanced. The material and methods section should be extended. Flowchart of methodology for the development of the algorithm looks interesting, but authors should add the section regarding image recognition technique as well as detailed description of algorithms (complexity, diagram representation).

Response: The part of ‘Material and method’ has been revised.

             The sentences have been added as follows;

In line 140-142; A binary image composed of black of clean surface with ‘0’ and white of residue with ‘1’ was generated by applying the thresholding value to the image with the selected waveband, and the regions in white were identified as residues.

In line 143-145; Fruit residues were identified, and accuracies of detection were calculated by the algorithms based on single waveband and two-waveband ratio.

3. Potential reader is not familiar with machine learning terminology. Authors should add the information regarding description of dataset, validation technique, accuracy measurement strategy and definition. E.g. were the same droplets used in training and testing stage? How final rules were identified. Did you use any of machine learning techniques (you mentioned about accuracy, validation, etc.)

Response: The sentences have been added in line 147-150 as follows;

Each data set is randomly distributed, and the optimal wavebands and accuracies are calculated from the calibration data. The selected wavebands were applied to the validation data to verify whether the analysis results are valid.

4. The experiment should be described in the way to allow the reader to repeat and validate results – the description of all parameters of the experiment, and each step should be included in the paper (e.g. MATLAB parameters, etc)

Response: All of the parameter such as preparation of samples, system configuration, excitation waveband, exposure time, method of data acquisition was described in Materials and methods section. So experiment is available to repeat.

5. The discussion can confuse the reader. What is the reason to provide long discussion regarding residue detection algorithms for different fruits? Are there serious difference between these algorithms? If so, please describe that in details. This will be more valuable for the reader than description of the figures. Please cut such description or improve abstract and title by identification of different types of fruits.

Response: It is the process of finding an algorithm that can be applied to various fruits at once since fruits have different spectral properties.

The sentence has been added in line 23 as follows;

Honeydew, orange, apple, and watermelon were selected as representatives since they are mainly used as fresh-cut fruits..

6. Authors mentioned, that on the figures ‘the average spectra’ for each fruits are presented, but there are no statistical analysis. Can you provide such analysis on the figures to justify your conclusions?

Response: The data analysis in this study uses the extracted pixel spectra to include the difference for each measurement location, so these spectra can represent the characteristics of the sample.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The manuscript has been improved according to previous comments of reviewers.

Author Response

We appreciate your comments

Reviewer 3 Report

Revised version of the manuscript entitled 'Development of fluorescence imaging technique to detect fresh-cut food organic residue on processing equipment surface' looks a little bit better than previous version, but is still far from the final version that can be accepted for publication.

Authors improved Figures - they are readable now, but still there are a lot of other weak points that were not corrected but important to prove and confirm authors' findings:

1. There is no comparison with other methods, but authors claimed that their methods is faster. By comparison one should understand results obtained from deseral method based on the same instances of the problem.
2. Authors mentioned about validation, accuracy - terms defined clearly in machine learning area, but there are no explanation how authors measured or calculated these results - case not acceptable for the scientific research
3. The description and analysis of algorithms applied/designed by authors is missing. The only one flowchart of the pipeline is not enough.
4. Authors mentioned about developed imaging algorithms, algorithms for detecting resudues, global algorithm, but there are no detailed description. It is impossible to evaluate them and estimate the correctness of them. There are no decription regarding MATHLAB for image processing, data extraction and analysis.
5. There are no statistical analysis for most of the presented results

Summing up, the description of the results and methodology presented by authors should be seriously improved. Unfortunatelly. the final recommendation for the current shape of the paper is 'Reject'.

Author Response

We appreciate your comments. The response to the review comments is as follows.

Authors improved Figures - they are readable now, but still there are a lot of other weak points that were not corrected but important to prove and confirm authors' findings:

Response : Supplymentary Figure 1-6 has been added.

1. There is no comparison with other methods, but authors claimed that their methods is faster. By comparison one should understand results obtained from deseral method based on the same instances of the problem.

Response : Line 50-65 has been changed

The sanitation monitoring of processing surfaces in food processing facilities is mostly a manual process (e.g., visual inspection) performed by inspectors according to the established guidelines [10]. The conventional sanitary inspection method, which is the microbial contamination inspection, is a method of determining contamination by selecting an area suspected of contamination, sampling substances on the surface of this area, culturing the microorganism for 1 to 3 days, and counting the number of microorganisms. This microbial culturing requires a laboratory process, and is time-consuming. Also nucleic-acid-based methods are more sensitive than traditional culture-based methods because they detect the DNA or RNA sequences of specific pathogens, but require skilled experts, time-consuming, and expensive equipment [11-12]. These methods can measure the degree of contamination differently depending on the selected sampling area, such as heavily contaminated point, normal contaminated point, or non-contaminated point, and the number of samples may be limited by the performance of the analysis equipment and the analysis time. The monitoring techniques that can be quickly measured without limiting the number of sampling and sampling locations has been required. Especially a system that can rapidly and accurately detect the presence of vectors for pathogen growth on the surfaces of food processing equipment is needed to enhance food safety [13].

Line 66-76 has been changed.

A variety of imaging and spectroscopic technologies have been researched for the rapid contamination detection in foods and food processing facilities [14-15]. In particular, hyperspectral imaging technology, which combines imaging and spectroscopy techniques, can promise nondestructive measurement technique related to the acquisition of spatial and spectral information simultaneously for each pixel in a sample image [16]. This technology can map differences in physical, chemical, and biological properties of a target object into a spatial distribution, and monitor the entire surface of the target object in real time. The technology has been studied in recent years for areas such as food safety inspection, quality identification of agricultural products, and biological contaminant detection [17], along with analysis techniques such as spectral unmixing, multivariate analysis-based band selection and classification for obtaining meaningful information from vast hyperspectral data.

2. Authors mentioned about validation, accuracy - terms defined clearly in machine learning area, but there are no explanation how authors measured or calculated these results - case not acceptable for the scientific research

Response : Line 135-173 were revised.

3. The description and analysis of algorithms applied/designed by authors is missing. The only one flowchart of the pipeline is not enough.

Response : Line 135-173 were revised.

4. Authors mentioned about developed imaging algorithms, algorithms for detecting resudues, global algorithm, but there are no detailed description. It is impossible to evaluate them and estimate the correctness of them. There are no decription regarding MATLAB for image processing, data extraction and analysis.

Response : Line 135-173 were revised and line 470-479 has explained imaging algorithms.

5. There are no statistical analysis for most of the presented results.

Response : One-way ANOVA method is one of the statistical analysis methods. Also, Supplymentary Figure 1-10 were added as One-way ANOVA analysis result ANOVA for classifying 2B and #4 finished Stainless steel surface and fruit residues.

Author Response File: Author Response.pdf