The Automated Detection of Fusarium Wilt on Phalaenopsis Using VIS-NIR and SWIR Hyperspectral Imaging

Round 1

Reviewer 1 Report

1. There is missing reference [1] in the manuscripts. It start with ref [2] line 41. Please check again the number for the reference

2. Page 3 Line 124 - there should be space between the unit and numbers : Example 162nm --> 162 nm. Please check throughout the manuscripts

3. Full abbreviation of BP comes after using the abbreviations. Page 7  Figure 4 .........a BP calculation.  & the full abbreviation is at Page 8

4. Please check the format of writing the references. Please standardize all the formatting

This paper need to send for English Proofreading.

Author Response

1. There is missing reference [1] in the manuscripts. It start with ref [2] line 41. Please check again the number for the reference

Line 39. Fusarium is one of the major diseases causing pathogens infecting orchids that is widely distributed in soil and associated with plants worldwide [1].

2. Page 3 Line 124 - there should be space between the unit and numbers : Example 162nm --> 162 nm. Please check throughout the manuscripts

We have completed a thorough review of the entire manuscript and made necessary corrections to the previously identified errors. In Section 2.2, we have also revised the presentation of the spectral camera specifications, using a concise table format for better clarity.

3. Full abbreviation of BP comes after using the abbreviations. Page 7  Figure 4 .........a BP calculation.  & the full abbreviation is at Page 8

Thank you for your reminder. Taking into account the reviewers' suggestions, we have decided to reduce the presentation of the band selection and band prioritization research. Instead, we will focus on the main objectives in writing this article. We have carefully reviewed the entire manuscript and made necessary revisions to the relevant sections accordingly.

4. Please check the format of writing the references. Please standardize all the formatting

Thank you for your reminder. We will review this part again.

Author Response File: Author Response.docx

Reviewer 2 Report

This paper proposed an automatic system for detection of Fusarium Wilt on Phalaenopsis based on VIS-NIR and SWIR hyperspectral imaging. According to the experimental studies, the proposed system can achieve 95.77% accuracy within a relatively short period of computing time, which makes it feasible in the industry. However, the reviewer had a few comments and suggestions as follows.

1. As for the technical contents, although this paper covered a bunch of algorithms and two hardware platforms, some key points were missing. First of all, it should be clearly explained the most critical issue this paper was trying to address. Assuming the pipeline identification system was the focus, then the “real time” might be the most important aspect. Second, how these algorithms contained in the paper were combined together to provide a solution to such an issue? What is the logic behind? This paper needs at least including an overall flowchart to explain each part in great detail. Third, what is the difference in data processing technique on two hardware platforms? Each hardware platform is supposed to have its own corresponding data processing flowchart for clear explanation.

2. As for the experimental studies, the classification and 3D ROC sections contain a large amount of numerical data. However, further discussion is required to explain the results. First of all, the studies showed that the PHMID provides better or equivalent performance compared with VNIR and SWIR hyperspectral sensors. Then, a fundamental question arises in the reviewer’s mind: why is it necessary to utilize much more expensive hyperspectral sensors for this topic? Second, a large number of numerical results tabulated at Table 9-11 need to be clearly explained. Third, the automation pipeline identification system section does not seem to be related to the experimental studies. According to the reviewer’s point of view, it should be moved to section 2. 

3. Overall speaking, the main purpose of this paper was to propose a data processing framework on VNIR/SWIR hyperspectral sensors and PHMID for detection of Fusarium Wilt on Phalaenopsis. Nevertheless, the innovative points of the proposed system and the importance of such a system have to be clearly stated. Besides, it would be better to bring a connection between VNIR/SWIR hyperspectral sensors and PHMID, which seem to be independent in this paper.

Extensive editing of English language is required

Author Response

This paper proposed an automatic system for detection of Fusarium Wilt on Phalaenopsis based on VIS-NIR and SWIR hyperspectral imaging. According to the experimental studies, the proposed system can achieve 95.77% accuracy within a relatively short period of computing time, which makes it feasible in the industry. However, the reviewer had a few comments and suggestions as follows.

1. As for the technical contents, although this paper covered a bunch of algorithms and two hardware platforms, some key points were missing. First of all, it should be clearly explained the most critical issue this paper was trying to address. Assuming the pipeline identification system was the focus, then the “real time” might be the most important aspect. Second, how these algorithms contained in the paper were combined together to provide a solution to such an issue? What is the logic behind? This paper needs at least including an overall flowchart to explain each part in great detail. Third, what is the difference in data processing technique on two hardware platforms? Each hardware platform is supposed to have its own corresponding data processing flowchart for clear explanation.

The current article has been revised to emphasize the pipeline identification system. The modifications were made from line 117 to line 232. The PHMID section was excluded from the revised article as it is not directly relevant to the main contributions of this study.

2. As for the experimental studies, the classification and 3D ROC sections contain a large amount of numerical data. However, further discussion is required to explain the results. First of all, the studies showed that the PHMID provides better or equivalent performance compared with VNIR and SWIR hyperspectral sensors. Then, a fundamental question arises in the reviewer’s mind: why is it necessary to utilize much more expensive hyperspectral sensors for this topic? Second, a large number of numerical results tabulated at Table 9-11 need to be clearly explained. Third, the automation pipeline identification system section does not seem to be related to the experimental studies. According to the reviewer’s point of view, it should be moved to section 2. 

Upon reevaluation of this article, PHMID was removed, and the focus was redirected towards the pipeline identification system composed of VNIR and SWIR. The differences between the high-spectral images and other studies were also readdressed in lines 59-83. Line 332-351 provided additional clarifications regarding the table. Moreover, certain content from the automation pipeline identification system section was transferred to section 2.1.

3. Overall speaking, the main purpose of this paper was to propose a data processing framework on VNIR/SWIR hyperspectral sensors and PHMID for detection of Fusarium Wilt on Phalaenopsis. Nevertheless, the innovative points of the proposed system and the importance of such a system have to be clearly stated. Besides, it would be better to bring a connection between VNIR/SWIR hyperspectral sensors and PHMID, which seem to be independent in this paper.

The innovation and system design of this article are discussed and explained in lines 399-451.

Author Response File: Author Response.docx

Reviewer 3 Report

The manuscript entitled "Automated Detection of Fusarium Wilt on Phalaenopsis Using VIS-NIR and SWIR Hyperspectral Imaging" introduces a unique automated detection system for Fusarium wilt disease in Phalaenopsis orchids, using hyperspectral imaging technology and various machine learning techniques to identify diseased plants accurately. The study aims to enhance detection, minimize equipment costs, and decrease computation time through feature conversion, band expansion, and band selection. The subject is undoubtedly crucial and relevant to the field. However, following an extensive review, I believe the manuscript requires considerable revision before consideration for publication. Below, I specify areas needing attention:

The introduction is missing ample context. A more thorough literature review and synthesis of existing knowledge would furnish better context and assist in articulating the study's unique contributions. Authors should provide references for each statement. For instance, in line 61 ("In recent years, hyperspectral technology has been widely used in detecting diseases of various crops"), authors should elaborate further and include references to works in which other authors have utilized images. Furthermore, the authors claim that "The advantage of hyperspectral image is that…" but they do not provide a background of works using RGB or multispectral images and why hyperspectral is superior. Authors should examine reviews of the topic to enhance their work. This review could aid the authors in better explaining their work concerning hyperspectral images: “Hyperspectral Sensing of Plant Diseases: Principle and Methods”. https://www.mdpi.com/2073-4395/12/6/1451

and this review could help the authors improve their explanation of their work regarding image use for disease detection: “Recent advances of application of optical imaging techniques for disease detection in fruits and vegetables: A review” https://doi.org/10.1016/j.foodcont.2023.109849.

Moreover, some sections appear repetitive. It would be beneficial to revise these sections to avoid redundancy. For instance, authors seem to have repeated information about the advantages of hyperspectral imaging and its application in agriculture in lines 51-60 and 61-71. Still, the authors almost did not include references to their statements.

Although the material and methods section is extensive, it lacks specific information for the study. Some methods, such as those involving hyperspectral image calibration (lines 147-153) and band selection (lines 178-186), require a more detailed explanation for clarity and reproducibility. Authors should include any used software, specific settings or parameters, and justifications for these choices. Authors should provide more information about the statistical methods used in the analyses, as well as the metrics used to evaluate their machine learning models. The section describing the simulated high temperature and humidity conditions for Fusarium growth (lines 108-110) lacks specific details. Please include the exact conditions used (temperature, humidity levels) to allow for experiment reproducibility. Also, the image sequencing is confusing. There are several specialized terms used throughout the text, including ATGP, SAM, CEM, VD, HFC, BS, and UBIS. While some of these are explained later in the text, initial instances should include a brief definition or full term name to ensure clarity.

Regarding the Figures and Tables, authors should cross-check the figures and table numbers to guarantee they correspond with the text references. For example, Figure 3 is referenced in line 161, but the description seems to match Figure 2. The quality of the figures needs to be enhanced. The resolution of Figure 6 is extremely low, and the letters and numbers within Figure 10 are not readable.

The results are reasonably clear, although the discussion is weak as the authors do not provide sufficient explanations of the impact of their findings and they do not compare them with the results of other authors. Authors should check the works already commented on. In addition, some conclusions could be better substantiated, which would better explain the findings of the work.

Lastly, the English language usage must be improved. Some examples:

·       "Phalaenopsis is one of the most welcome flowers by people" - incorrect use of "welcome".

·       "has highly economic and ornamental value" - incorrect adjective usage.

·       "These spp. has severely reduced" - incorrect subject-verb agreement. (It should be: "These spp. have severely reduced").

·       "including Support Vector Machine (SVM), deep neural network (DNN) and Random Forest Classifier, then use 3D Receiver Operating Feature (3D ROC) to evaluate the effectiveness of those models" - incorrect tense. It should be “used”

·       "Hypersepctral" in line 138.

I recommend that the authors thoroughly proofread the entire manuscript as numerous English language errors are present.

Specific comments

Line 61-70

References should be added

Line 63

“The advantage of hyperspectral image is that it can reserve the entire electromagnetic spectrum of each pixel which usually consists of a visible spectrum and an invisible spectrum.”

The advantage compared to what sensors? RGB? Multispectral?

The authors should elaborate on the advantage of hyperspectral images in comparison to other sensors like RGB or Multispectral. To improve readability, the authors can indicate some works related to multispectral and RGB images to detect diseases:

“On the Potentiality of UAV Multispectral Imagery to Detect Flavescence dorée and Grapevine Trunk Diseases”. https://doi.org/10.3390/rs11010023

“Mapping the Spatial Variability of Botrytis Bunch Rot Risk in Vineyards Using UAV Multispectral Imagery.” https://doi.org/10.1016/j.eja.2022.126691

“Comparison of Canopy Shape and Vegetation Indices of Citrus Trees Derived from UAV Multispectral Images for Characterization of Citrus Greening Disease.” https://doi.org/10.3390/rs12244122

and then, the authors can suggest what is the benefit of using hyperspectral images compared to these works.

Line 71

The statement "can be detected in real time" requires more clarity. How does it compare to the speed of other sensors such as RGB or MS, which are generally faster?

Line 106

Additional context regarding the selection of Phalaenopsis Sogo Yukidian 'V3' for this study would be beneficial.

Lines 104-113

More comprehensive information on the selection criteria for the plantlets is necessary. Moreover, clarification about the environmental conditions simulated for Fusarium growth is needed.

Line 113

The claim that plantlets were obtained in August 2021, but data filmed in April 2021 was also used, seems conflicting. Please provide more clarity on the timelines involved in the data collection process.

Line 115-116

The phrase "after a period" is vague. Could you specify the duration involved?

Lines 121-126

Additional details about the configuration of the hyperspectral cameras, specifically the reasoning behind the selected spectral ranges and spatial line pixels for VNIR and SWIR hyperspectrometers, are necessary. Also, what specific features do the two halogen lamps used in the experiment have?

Lines 131-138

More detailed information about how the PHMID was utilized in this study is needed. Further, an explanation behind the selection of spectral ranges for the LEDs would be beneficial.

Lines 147-153

The formula in the hyperspectral image calibration section lacks transparency. Please provide a comprehensive explanation for the symbols in the formula to enhance understanding.

Lines 154-161

An in-depth explanation of the ATGP method and the process of how CEM processing was executed to obtain the final ROI would be beneficial.

Lines 178-186

Providing more details on the band selection process, especially how statistical criteria were applied to identify the most effective bands, would be helpful.

Lines 188-199

In the band de-correlation section, it would be valuable if the authors could provide further explanation about the process and criteria for band prioritization.

Lines 207-209

The Band Priority section would benefit from specifying the statistical criteria used and more information about the calculations.

Lines 210-223

More details on how autocorrelation and cross-correlation were implemented in the Band Expansion Process section are required.

Lines 232-237

Additional information on how DNN, SVM, and RF were employed in this specific study would be beneficial, particularly regarding their application in the data training and prediction processes.

Line 419-426

At the start of the discussion, the authors describe the disease progression in phalaenopsis orchids. It would be helpful to provide more specific context about how their research contributes to advancements in this area. Moreover, clarity on the relationship between the number of diseased pixels, the progression of time, and the increasing accuracy rate would enhance this section.

Line 427-435

The claim that the mean-variance method "doubles the number of features" (line 433) lacks clarity. It would be useful if the authors could elaborate further on this point.

Line 440-452

While the authors offer good descriptions of the user interface of the Automation Pipeline Identification System, providing more analysis on how this system might influence user experience and its potential practical implications would be beneficial.

Line 453-465

The link between the conclusions and the broader objectives of the study seems somewhat unclear. Detailed explanations on how these findings contribute to the field, their implications, and potential future research directions would enhance this section.

Line 466

Providing some context for the mentioned acceptable range of detection time could be useful. Is this range based on industry standards or user expectations?

Lastly, the English language usage must be improved. Some examples:

·       "Phalaenopsis is one of the most welcome flowers by people" - incorrect use of "welcome".

·       "has highly economic and ornamental value" - incorrect adjective usage.

·       "These spp. has severely reduced" - incorrect subject-verb agreement. (It should be: "These spp. have severely reduced").

·       "including Support Vector Machine (SVM), deep neural network (DNN) and Random Forest Classifier, then use 3D Receiver Operating Feature (3D ROC) to evaluate the effectiveness of those models" - incorrect tense. It should be “used”

·       "Hypersepctral" in line 138.

I recommend that the authors thoroughly proofread the entire manuscript as numerous English language errors are present.

Author Response

The manuscript entitled "Automated Detection of Fusarium Wilt on Phalaenopsis Using VIS-NIR and SWIR Hyperspectral Imaging" introduces a unique automated detection system for Fusarium wilt disease in Phalaenopsis orchids, using hyperspectral imaging technology and various machine learning techniques to identify diseased plants accurately. The study aims to enhance detection, minimize equipment costs, and decrease computation time through feature conversion, band expansion, and band selection. The subject is undoubtedly crucial and relevant to the field. However, following an extensive review, I believe the manuscript requires considerable revision before consideration for publication. Below, I specify areas needing attention:

The introduction is missing ample context. A more thorough literature review and synthesis of existing knowledge would furnish better context and assist in articulating the study's unique contributions. Authors should provide references for each statement. For instance, in line 61 ("In recent years, hyperspectral technology has been widely used in detecting diseases of various crops"), authors should elaborate further and include references to works in which other authors have utilized images. Furthermore, the authors claim that "The advantage of hyperspectral image is that…" but they do not provide a background of works using RGB or multispectral images and why hyperspectral is superior. Authors should examine reviews of the topic to enhance their work. This review could aid the authors in better explaining their work concerning hyperspectral images: “Hyperspectral Sensing of Plant Diseases: Principle and Methods”. https://www.mdpi.com/2073-4395/12/6/1451

and this review could help the authors improve their explanation of their work regarding image use for disease detection: “Recent advances of application of optical imaging techniques for disease detection in fruits and vegetables: A review” https://doi.org/10.1016/j.foodcont.2023.109849.

Moreover, some sections appear repetitive. It would be beneficial to revise these sections to avoid redundancy. For instance, authors seem to have repeated information about the advantages of hyperspectral imaging and its application in agriculture in lines 51-60 and 61-71. Still, the authors almost did not include references to their statements.

Modifications have been made in lines 59-83.

Although the material and methods section is extensive, it lacks specific information for the study. Some methods, such as those involving hyperspectral image calibration (lines 147-153) and band selection (lines 178-186), require a more detailed explanation for clarity and reproducibility. Authors should include any used software, specific settings or parameters, and justifications for these choices. Authors should provide more information about the statistical methods used in the analyses, as well as the metrics used to evaluate their machine learning models. The section describing the simulated high temperature and humidity conditions for Fusarium growth (lines 108-110) lacks specific details. Please include the exact conditions used (temperature, humidity levels) to allow for experiment reproducibility. Also, the image sequencing is confusing. There are several specialized terms used throughout the text, including ATGP, SAM, CEM, VD, HFC, BS, and UBIS. While some of these are explained later in the text, initial instances should include a brief definition or full term name to ensure clarity.

The band selection content has been removed due to the emphasis on the automated platform in the modified version. Detailed sample placement conditions and image processing are presented in lines 140-200.

Regarding the Figures and Tables, authors should cross-check the figures and table numbers to guarantee they correspond with the text references. For example, Figure 3 is referenced in line 161, but the description seems to match Figure 2. The quality of the figures needs to be enhanced. The resolution of Figure 6 is extremely low, and the letters and numbers within Figure 10 are not readable.

Thank you for the reminder. The tables and figures have been reviewed again, and unclear figures have been replaced with tables for clarity.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The reviewer acknowledges the efforts of the authors to improve the quality of this paper significantly. Basically, most of the major concerns have been addressed. There are still a few minors required to be addressed.

1. Even though the introduction section had been improved. The reviewer suggested including a summary of the contributions of the proposed system to the end of the section, instead of going through the flowchart of the proposed system, which should be the major focus of the next section.

2. Although section 2.6 was called “Feature Conversion”, there is no linear or nonlinear conversion involved. Instead, the proposed method is statistical indicators within a window, such as mean and variance. The statistical indicators were joined with the original pixel to enhance the performance of classifiers. As a result, the reviewer suggests modifying the name of this section. 

3. Table 7-8 tabulates a large amount of numerical results of 3D ROC. Unfortunately, the explanation below was not able to guide the readers to go through the content. The authors might consider reducing the content of tables to some importantly related indicators, such as PD, PF, SNPR. Since the SNPR index seems to be the only indicator to determine the superior performance of DNN by the authors.

4. Although the authors submitted the records of track changes for clear indication of modification of this revision. However, it is difficult to read in some parts, such as page 3, 11, 16. Besides, some typos might need to be carefully corrected, such as the titles of section 2.8.1, “D-Receiver Operating Characteristic Curve Analysis(3D-ROC)”.

Overall, the quality of English was improved significantly, but there are some typos might need to be carefully corrected.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

The revised version of the paper titled "Automated Detection of Fusarium Wilt on Phalaenopsis Using VIS-NIR and SWIR Hyperspectral Imaging" has been enhanced, with a majority of the reviewers' feedback integrated. Nevertheless, there are a few comments, predominantly concerning figures and tables, that still need to be addressed before publication.

In general, figure and table captions must be self-explanatory. Therefore, improvements are required throughout the manuscript in this aspect. For instance, table 8 does not define the acronyms used, such as H, D, PD_τ, PF_τ, etc., in its caption. Likewise, the caption of Figure 1 is overly brief and fails to provide sufficient detail about the content of the figure. The caption should be improved, for instance, as “Hyperspectral imaging system composed of a hyperspectral camera and halogen lamp mounted over a conveyor belt...". Similar revisions are needed for all figure captions within the manuscript.

Moreover, in general, a higher quality of the figures is necessary. The text in Figure 7, for example, is currently unreadable.

Lastly, adherence to scientific writing norms is essential. In compliance with binomial nomenclature, the genus and species of organisms should be italicized (Fusarium, Phalaenopsis…). Thus, the entire manuscript must be reviewed accordingly. A useful example of the correct formatting can be found in one of the paper's own references:

“Develop an efficient inoculation technique for Fusarium solani isolate ‘TJP-2178-10’ pathogeny assessment 519 in Phalaenopsis orchids. https://doi.org/10.1186/s40529-021-00310-z"

Author Response

Please see the attachment.

Author Response File: Author Response.docx