Grading and Detection Method of Asparagus Stem Blight Based on Hyperspectral Imaging of Asparagus Crowns

Round 1

Reviewer 1 Report

 As a reader, It is not clear to me what organism causes asparagus stem blight. It is not clear under what settings the experiment was conducted (growth chamber, greenhouse, or field conditions) and not clear how the disease severity was measured (before the image analysis), and whether the image-derived severity rates were highly correlated with human-assessed values with a high statistical power or not. My specific comments can be found in the attached MS. 

Comments for author File: Comments.pdf

Please get help from proof reading services.

Author Response

Review comments:

As a reader, It is not clear to me what organism causes asparagus stem blight. It is not clear under what settings the experiment was conducted (growth chamber, greenhouse, or field conditions) and not clear how the disease severity was measured (before the image analysis), and whether the image-derived severity rates were highly correlated with human-assessed values with a high statistical power or not. My specific comments can be found in the attached MS. 

 

Reply to comments:

• Asparagus stem blight is caused by Phomopsis asparagi (Sacc.), which widely survives in nature and can be widely spread under suitable conditions.
• The experiment was carried out in a greenhouse environment, and asparagus plants were naturally infected with stem blight. The night temperatures was 22 ~ 26℃, and the day temperatures was 28 ~ 35℃ in the greenhouse where the asparagus plants
• In this study, the disease grade of asparagus stem blight of mother stem was divided by manual assessment. The plants without stem blight symptoms both on the canopy and the stem of mother stem plants were healthy plants, and the disease grade was ’1’. Asparagus plants without stem blight symptoms on the canopy but with stem blight symptoms on the stem of mother stem plants were mildly affected plants, and the disease grade was' 2 '. Asparagus plants with stem blight symptoms on both the canopy and the stem of the mother stem plant were moderately and severely affected plants, and the disease grade was' 3 '.
• In this study, asparagus stem blight was naturally infected in a greenhouse environment, therefore, the numbers of asparagus plants in different disease grades were different.
• Asparagus stem blight mainly harms the stem, and the thick canopy of mother stem easily cover the symptoms of stem blight disease of the stem. To assess the severity of the disease manually, it is usually necessary to pull away the canopy of the mother stem to observe the disease degree of stem. However, manually assessing the severity of stem blight only through the canopy makes it difficult to distinguish the healthy plants of grade '1' from the mild affected plants of grade '2'. The observation of the moderate and severe diseased plants of grade '3' indicates that the plants have been seriously affected, and the disease is difficult to control at this time. In this study, the hyperspectral imaging technology of asparagus mother stem canopy combined with machine learning algorithm was explored to detect the disease degree of asparagus stem blight, and the results were similar to those of manual evaluation as shown in table 1~3.
• In the spectral data preprocessing methods ,First derivative (FD)          preprocessing of spectral data can effectively resolve overlapping peaks。

Author Response File: Author Response.docx

Reviewer 2 Report

see the attachment 

Comments for author File: Comments.pdf

 Minor editing of English language required

Author Response

1.The details of the dataset are not provided in the abstract. It would be beneficial for readers if the author could include essential information, such as the number of images, the number of classes, and their size (in terms of images). Providing these details would give readers a better understanding of the dataset's complexity and help in assessing the model's performance comprehensively.

Re: There were 503 sets of hyperspectral data in training set containing 165, 150, 188 sets of data of samples in disease grade 1, grade 2 and grade 3 in sequence, and there were 167 sets of hyperspectral data in test set containing 55, 50, 62 sets of data of samples in disease grade 1, grade 2 and grade 3 in sequence.

 

2. The author has presented the results in the abstract, which is commendable. However, to gain a more comprehensive view of the model's effectiveness, it would be valuable to include comparison results with other existing models. By showcasing how the proposed model performs in comparison to state-of-the-art methods, readers can better appreciate its novelty and strengths. Additionally, it would be beneficial to provide the value of the loss function employed during training. This would give readers insights into the model's convergence and how well it fits the data.

Re: The value of the loss function employed during training has been provided in the abstract, Table 2~ Table 13 and the main text.

 

3.Contribution missing: the explicit statement of the article's contribution. To enhance the introduction further, it is essential for the author to provide a clear and concise overview of the article's contributions. This can be achieved by adding a dedicated section towards the end of the introduction, presenting the contributions in bullet points.

Re: The statement of the article's contribution has been added at the end of the introduction.

 

4. Introduction contains limited information when it comes to literature. Author must add more recent works about deep learning in agriculture. Some of the studies are as follows, author should include those in literature and highlight their contribution in the literature. Fruit Image Classification Model Based on MobileNetV2 with Deep Transfer Learning Technique; Enhancing Image Annotation Technique of Fruit Classification Using a Deep Learning Approach; Image Acquisition, Preprocessing and Classification of Citrus Fruit Diseases: A Systematic Literature Review; Harnessing the Power of Transfer Learning in Sunflower Disease Detection: A Comparative Study; Classification of hazelnut kernels with deep learning.

Re: Recent works about deep learning in agriculture have been added in Introduction section.

 

5. Section 2.3: The author has presented a comprehensive discussion of Hyperspectral imaging; however, it would be immensely beneficial for readers if the author could include a sample Hyperspectral image in this section. Adding an illustrative example would greatly enhance the understanding of the concepts discussed. - Dataset details lack clarity, particularly regarding the number of images per class. To address this, the author is encouraged to present the dataset information in a structured and accessible format, such as a tabular form. By doing so, readers can easily grasp the distribution of images across different classes, making the research more transparent and replicable.

Re: Sample Hyperspectral images have been included in section 2.3, and dataset information has been provided in a table in section 2.3.

 

6. It is crucial for any research to include evaluation matrices to assess the performance of the proposed method rigorously. In this regard, the author should consider incorporating evaluation metrics into the study. Commonly used metrics such as accuracy, precision, recall, F1-score, or confusion matrix would provide valuable insights into the model's effectiveness and enable comparison with other approaches in the field. By including these evaluation metrics, the paper's scientific rigor and credibility would be significantly improved.

Re: Evaluation metrics of accuracy, precision, recall, F1-score, and confusion matrix have been incorporated in this study.

 

7. In Figure 5, 6, the absence of labels for the graphs makes it challenging for readers to discern the intended message. To enhance understanding, it is essential for the author to include clear and informative labels for each graph.

Re: Informative labels have been included In Figure 5, 6.

 

8. L1, L2 and L3 is not defined prior to Figure 7.

Re: L1、L2 and L3 have been changed into Grade 1, Grade 2, Grade 3.

 

9. The comparison of different preprocessing methods for the discrimination model is informative. However, it would be beneficial to provide a concise summary or table that presents the discrimination accuracies for each model, making it easier for readers to grasp the key findings at a glance.

Re: It has been modified as requested in the main text.

 

10. In Figure 10, which illustrates the influence of the K value on the KNN performance, it would be valuable to add specific data points or labels on the graph to indicate the exact discrimination accuracies for each K value. This will allow readers to visualize the results more effectively and draw conclusions based on the data presented.

Re: Specific data points and labels have been added in Figure 10.

 

11. Section 3.5: Clarify SG Smoothing Effect: In the first point of the discussion, you mention that the SG smoothing preprocessing method had no significant effect on improving the accuracy of the disease grade discrimination model. It would be helpful to elaborate on why this is the case. Providing more details about the elimination of noise interference when averaging the reflection spectrum in the region of interest would help readers understand the rationale behind this observation.

Re: A figure has been provided to explain the reason why the SG smoothing preprocessing method had no significant effect on improving the accuracy of the disease grade discrimination model。

 

12. Data Comparison: To strengthen your findings, consider including a comparative analysis of different preprocessing methods, including the benefits and limitations of each approach. This could aid readers in understanding why certain methods were more effective than others and provide a more comprehensive understanding of the model's performance.

Re: The comparative analysis of different preprocessing methods has been carried out together with the comparative analysis of different model effects, and the benefits and limitations of each approach has been included in 2.4 Section.

 

13. Clarify Optimal Modeling and Preprocessing Methods: In the second point, you mentioned that optimal modeling and preprocessing methods were found. Please elaborate on the specific reasons why these methods were deemed optimal and how they align with the requirements of lightweight models. This will help readers better understand the reasoning behind your choices and the potential advantages of using these methods in similar scenarios.

Re: It has already been explained in the second point of the discussion section.

 

14.Discuss Future Research Directions: Considering the feasibility of hyperspectral detection of stem blight based on asparagus mother stem canopies, it would be beneficial to discuss potential avenues for future research in this domain. Addressing possible areas of further investigation and the potential impact of the findings on the development of corresponding equipment will add depth to the discussion

Re: Directions for future research have been added to the discussion section according to your suggestions

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

author has responded all the comments. 

Minor editing of English language required