Proposed Fuzzy-Stranded-Neural Network Model That Utilizes IoT Plant-Level Sensory Monitoring and Distributed Services for the Early Detection of Downy Mildew in Viticulture
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis paper proposes a novel monitoring architecture for early detection of downy mildew in grape cultivation. It combines Internet of Things (IoT) technology and semi-supervised deep learning algorithms to enhance disease detection. The study focuses on the "debina" grape variety and aims to improve the accuracy and efficiency of disease monitoring in vineyards. The methodology and experimental design involve the integration of IoT devices to collect data, which is then processed through deep learning models.
Comment 1: The introduction could more clearly explain the research gap and how this study aims to fill it. This article focuses on the detection of downy mildew in viticulture and can more clearly explain the unique contribution of this study.
Comment 2: What are the specific parameters measured by IoT sensors? How to detect a mold outbreak? By observing the state of plants? It can shed light on how deep learning models process and utilize data collected by IoT devices.
Comment 3: The article uses semi-supervised deep learning algorithm and fuzzy annotation. Further advantages of this algorithm for early detection can be provided.
Comment 4: The leaf area index for plant detection can be measured using RGB cameras such as drones, and hyperspectral imaging can further capture the red edge effect of plants, which is an important indicator for monitoring plant growth status. Therefore, it is recommended to introduce a discussion on hyperspectral imaging. For example, "A Gaze Video Rate High-Throughput Hyperspectral Imaging System and Its Application in Biological Sample Sensing" proposes a discussion of hyperspectral systems.
Comment 5: What specific methods are used to verify the accuracy of the deep learning model? Are there any comparative studies with traditional detection methods?
Comment 6: The conclusion could more directly summarize the main findings and their implications for viticulture. Additionally, potential limitations of this study and areas for future research can be highlighted.
Comments on the Quality of English LanguageThe use of English is usually very good with few errors. However, in order to improve the readability of the paper and ensure consistency throughout the entire process, I suggest conducting small-scale review or proofreading.
Author Response
This paper proposes a novel monitoring architecture for early detection of downy mildew in grape cultivation. It combines Internet of Things (IoT) technology and semi-supervised deep learning algorithms to enhance disease detection. The study focuses on the "debina" grape variety and aims to improve the accuracy and efficiency of disease monitoring in vineyards. The methodology and experimental design involve the integration of IoT devices to collect data, which is then processed through deep learning models.
Response: Thank you for your time and effort in reviewing our manuscript. Here, we quote our responses and amendments performed based on your comments.
Comment 1: What are the specific parameters measured by IoT sensors? How to detect a mold outbreak? By observing the state of plants? It can shed light on how deep learning models process and utilize data collected by IoT devices.
Response: In our experimental scenario, the IoT motes' sensors are temperature and humidity sensors. The thingsAI system motes support different profile types, as amended in Line 380. The first paragraph of the Experimental Scenario section has been amended to indicate the motes, meteo stations used, viticulture field placements, and their equipped sensors.
Comment 2: The article uses semi-supervised deep learning algorithm and fuzzy annotation. Further advantages of this algorithm for early detection can be provided.
Response: Two additional paragraphs have been added in section 2.4 (first and second) to elaborate on the term early and the use of fuzzy logic to provide autoencoded annotations.
Comment 3: The leaf area index for plant detection can be measured using RGB cameras such as drones, and hyperspectral imaging can further capture the red edge effect of plants, which is an important indicator for monitoring plant growth status. Therefore, it is recommended to introduce a discussion on hyperspectral imaging. For example, "A Gaze Video Rate High-Throughput Hyperspectral Imaging System and Its Application in Biological Sample Sensing" proposes a discussion of hyperspectral systems.
Response: Thank you for the recommendation. Section 1.3, paragraph 12 (from the start) adds an additional paragraph and three references for HSI systems.
Comment 4. What specific methods are used to verify the accuracy of the deep learning model? Are there any comparative studies with traditional detection methods?
Response: Thank you for asking. Unfortunately, there are no comparative studies of different models in the literature. Usually, proposed mechanistic models are formulated using fungus traps in the vine fields and measuring the number of oospores captured. In papers where modern DSS systems utilize vine-level sensors, the most simplified models are used, such as the Goidanich, the UR, EPI or DMCast models.
Additional deep learning model validation information has been added in section 3.4 with cross-comparison results of our model with the Goidanich model for 2023 of mildew outbursts in the Zitsa area.
Comment 5: The conclusion could more directly summarize the main findings and their implications for viticulture. Additionally, potential limitations of this study and areas for future research can be highlighted.
Response: The conclusion section has been rewritten to focus on the advantages for the viticulture Industry of using a dense sensor grid that utilizes open-source software and IoT, such as thingsAI. Also, the authors highlight the implications of models in viticulture that are easy to train by fuzzy annotating sensory measurements. The data annotation process can be automated using autoencoders to provide vine-level big-data collections as trainable datasets based on human-understandable fuzzy linguistic rules and fuzzy logic. Such data can be exploited to train deep learning models for precise classification and prediction purposes rather than deterministic mechanistic models that are hard to implement. Future research and potential limitations have been added to the conclusions.
Comment 6: The use of English is usually very good with few errors. However, in order to improve the readability of the paper and ensure consistency throughout the entire process, I suggest conducting small-scale review or proofreading.
Response: The authors proofread their manuscript, correcting typos, errors, and hard-to-read sentences to improve its readability. Several minor changes throughout our manuscript have been made to improve the readability.
Reviewer 2 Report
Comments and Suggestions for AuthorsThe following are my comments for improvement of this paper:
1. Several fact-based statements throughout the paper are missing supporting references. For instance, the statement – “Such technologies refer to sensors than monitor environmental conditions of a wide area of interest to technologies that monitor the conditions close to a single plant or a cluster of plants” should have a supporting reference.
2. Multiple typos throughout the paper. For example – “The implemented sensory devices called motes design is illustrated in Figure ??”. Here, instead of ?? a Figure number should be present.
3. The flowchart for the LoRaWAN class (Figure 3) does not have an “End” block. If the system architecture is designed to function as a continuous feedback system then an “End” block is not necessary. If not, then an “End” block should be present.
4. A comparison with prior works is missing: Please include a comparative study (qualitative and quantitative) with prior works in this field to highlight the novelty of this work.
5. The beginning of the Introduction section should be rewritten. The authors state – “Plasmopara viticola (P. viticola) is a North American origin fungus that invaded Europe”. This is true but it is very superficially stated. This statement should be supported with additional information and facts to highlight the need for investigation/research work in this area.
6. The limitations of the work should be clearly stated.
Author Response
Thank you for your time and effort in reviewing our manuscript. Here, we quote our responses and amendments performed based on your comments.
Comment 1: Several fact-based statements throughout the paper are missing supporting references. For instance, the statement – "Such technologies refer to sensors than monitor environmental conditions of a wide area of interest to technologies that monitor the conditions close to a single plant or a cluster of plants" should have a supporting reference.
Response: Supporting references have been added to the paragraph at line 51. Supporting references have also been added on lines 57 and 61. Also, on line 64. Supporting references have been added throughout our manuscript whenever needed.
Comment 2: Multiple typos throughout the paper. For example – "The implemented sensory devices called motes design is illustrated in Figure ??". Here, instead of ?? a Figure number should be present.
Response: Thank you very much for your remark. It was a reference (fig:1a)-labeling (fig1:a) issue and has been corrected. The authors also checked every image and table reference throughout their manuscript. The authors proofread their manuscript, correcting typos, errors, and hard-to-read sentences to improve its readability.
Comment 3: The flowchart for the LoRaWAN class (Figure 3) does not have an "End" block. If the system architecture is designed to function as a continuous feedback system then an "End" block is not necessary. If not, then an "End" block should be present.
Response: Figure 3. The flowchart has been amended to indicate a periodic probing process, moving from the sleep state to start probing for sensory measurements and transmitting them using LoRaWAN class A protocol.
Comment 4: A comparison with prior works is missing: Please include a comparative study (qualitative and quantitative) with prior works in this field to highlight the novelty of this work.at line 298, where "wither" should be replaced with "either."
Response: Prior works have been included, and the paragraph has been modified on line 298. Additionally, a paragraph has been added below Table 3, indicating both Senviro and ThingsBoard platforms that cover necessary DSS capabilities as mentioned by the authors as Key Performance Indicators for DSS systems.
Comment 5: The beginning of the Introduction section should be rewritten. The authors state – "Plasmopara viticola (P. viticola) is a North American origin fungus that invaded Europe". This is true but it is very superficially stated. This statement should be supported with additional information and facts to highlight the need for investigation/research work in this area.
Response: The beginning paragraph of the Introduction section has been rewritten. It Describes the fungus stages and contains information highlighting the need for investigation in this area.
Comment 6: The limitations of the work should be clearly stated.
Response: Limitations of this work have been added to the fifth paragraph of the conclusions section. Furthermore, several minor changes throughout our manuscript have been made to improve the readability.
Round 2
Reviewer 2 Report
Comments and Suggestions for AuthorsThe authors have revised their paper as per all my comments and feedback. I do not have any additional comments at this point. I recommend the publication of this paper in its current form.
Author Response
Thank you very much for your keen effort and comments, which helped us improve our manuscript.
On behalf of the authors' team
Best regards
Sotirios Kontogiannis
