Enhancing Corn Pest and Disease Recognition through Deep Learning: A Comprehensive Analysis
Round 1
Reviewer 1 Report
The authors recognized maize leaf blight, Helminthosporium maydis, gray leaf spot, rust disease, stem borer, and corn armyworm. The recognition accuracy reaching 96.02% was achieved. However, the authors only distinguished corn pests and diseases. Why did the dataset not include also healthy samples?
line 17: Please add for which model this accuracy was achieved.
The study was performed for publicly available datasets. The authors obtained high accuracy for the developed models. After that, did the authors try to apply these models to other independent samples from other places or seasons?
The discussion should be expanded and supplemented with references.
Author Response
This letter is sent to answers all question of Reviewer 1.
The authors recognized maize leaf blight, Helminthosporium maydis, gray leaf spot, rust disease, stem borer, and corn armyworm. The recognition accuracy reaching 96.02% was achieved. However, the authors only distinguished corn pests and diseases. Why did the dataset not include also healthy samples?
R.: In order to train the software to recognize diseases, it was necessary to give the software a standard of healthy images, which in this specific case was 500. We apologize to the reviewer for not including this detail, as we believed it would not be necessary, but we accept the reviewer's suggestion and included the software validation test.
line 17: Please add for which model this accuracy was achieved.
R.: This accuracy is achieved by the Resnet50 model with the addition of the ECA attention mechanism.
The study was performed for publicly available datasets. The authors obtained high accuracy for the developed models. After that, did the authors try to apply these models to other independent samples from other places or seasons?
R.: We are considering applying these models to other independent samples from other places or seasons and have designed a field robot that can automatically collect data. However, the specific results still need further experimental verification.
The discussion should be expanded and supplemented with references.
R.: We strongly agree with the points you have raised and have made corresponding changes in the paper and added references.
Reviewer 2 Report
1. The title of the paper is not consistent with the objective and methodology of the paper.
I suggest modifying it
2. Your results need more discussion.
3. You have to check the paper grammatically.
4. Punctuation.
5. The arrangement of the paper is not good.
6. You have to study more significant previous studies.
7. You have to focus on the analysis of your results.
8. You should put a section for mathematical analysis of the paper.
9. The paper needs a more modern reference.
10. The paper has some corrections and modifications.
11. The paper has a lot of printing errors and corrections.
12. You mention "Deep learning models, such as AlexNet, VGG, DenseNet, and ResNet,". How to compare them?
13. Explain the training strategy which you used in this paper for experiments. In a sub-section.
14. Equation (1) is not clear, You said " where x is the constant mapping; F(x) is the residual mapping; H(x) is the unknown mapping. The residual module is shown in Figure 4".
15. How do you interpret x as the constant mapping. Explain The functions F(x) and H(x) mathematically.
16. The figures should be centralized.
17. Use only HW not HxW in Equation (2).
18. Correct "3.2.2. Proposed Model ". It should be in new section and Model building in subsection.
19. You should refer to the references of the mathematical equations in Equations (1)-(4).
20. What's the relation between Adam Optimizer and the proposed model.
21. Put an algorithm in some steps for Model building.
22. What do you mean by "4.1. Subsection"?
23. How the numerical results are calculated and evaluated in Table 1.
24. Every equation should end with "," or ".".
Comments for author File: 
Comments.pdf
Moderate editing of English language required
Author Response
This letter is sent to answers all question of Reviewer 2.
1.The title of the paper is not consistent with the objective and methodology of the paper. I suggest modifying it
R.: The tittle was changed to best represent our objectives
- Your results need more discussion.
R: Thank you for your valuable suggestion the results were now briefly discussed with updated references. Please see the discussion section of this revised version.
- You have to check the paper grammatically.
R.: Thank you for your suggestions. This dissertation was purchased with MDPI's English touch-up service.
- Punctuation.
R.: Thank you for your suggestions. This dissertation was purchased with MDPI's English touch-up service.
- The arrangement of the paper is not good.
R.: We appreciate your feedback. We have restructured the paper to improve its overall organization and flow in a clear and logical manner.
- You have to study more significant previous studies.
R.: Thank you for the suggestion. We have conducted a comprehensive review of relevant literature and incorporated it in the discussion section to provide a more thorough context for our research.
- You have to focus on the analysis of your results.
R.: Thank you for your suggestion. We understand the importance of result analysis. We have provided a detailed analysis of our experimental results, including ablation experiments, comparisons with other models, confusion matrix evaluation, and performance visualizations.
- You should put a section for mathematical analysis of the paper.
R.: Thank you for your suggestions. We added a mathematical analysis of Adam's optimizer in Subsection 3.2.3.
- The paper needs a more modern reference.
R.: Thank you for your suggestion. We have updated the reference section with most recent and relevant sources to support our research.
- The paper has some corrections and modifications.
R.: We have carefully reviewed the manuscript and made necessary corrections and modifications.
- The paper has a lot of printing errors and corrections.
R.: We apologize for any printing errors or corrections in the initial submission. We have thoroughly proofread the paper and ensured that all printing errors have been rectified.
- You mention "Deep learning models, such as AlexNet, VGG, DenseNet, and ResNet,". How to compare them?
R.: To compare these deep learning models, we conducted experiments to compare the accuracy curves and training loss curves of several deep learning models. The best experimental result is ResNet, followed by DenseNet, and finally VGG and AlexNet.
- Explain the training strategy which you used in this paper for experiments. In a sub-section.
R.: Thank you for your suggestions. We have elaborated in subsection 3.2.
- Equation (1) is not clear, You said " where x is the constant mapping; F(x) is the residual mapping; H(x) is the unknown mapping. The residual module is shown in Figure 4".
R.: where: x is the output value, which is added to the constant mapping of the input x while entering the second weight layer through the ReLU activation function; F(x) is the residual mapping and H(x) is the ideal mapping.
- How do you interpret x as the constant mapping. Explain The functions F(x) and H(x) mathematically.
R.: The core of the ResNet model lies in proposing a shortcut connection, as shown in Fig. 4, where the output value x enters the first weight layer to get to the residual mapping F(x); while entering the second weight layer through the ReLU activation function, a constant mapping of the input x is added to obtain the ideal mapping F(x) +x. The advantage of the residual structure is to make the transfer between the shallow input value x and F(x) +x more sensitive.
- The figures should be centralized.
R.: With the exception of Figure 4, all others are centered. We believe the error happened after submission and submission to the reviewer. In any case, we double-checked each of the figures. Thank you
- Use only HW not HxW in Equation (2).
R.: Done, thank you for your improvement
- Correct "3.2.2. Proposed Model ". It should be in new section and Model building in subsection.
R.: Done, thank you for your improvement
- You should refer to the references of the mathematical equations in Equations (1)-(4).
R.: We have added references [37-38], thank you for your improvement.
- What's the relation between Adam Optimizer and the proposed model.
R.: The Adam optimizer can adapt the learning rate according to different parameters when the loss gradient is updated. Based on the ResNet50 model, the fusion of the channel attention mechanism and the Adam optimizer can improve the feature extraction ability of the model for maize blade pest and disease images.
- Put an algorithm in some steps for Model building.
R.: Thank you for your suggestion. We have added detailed steps of algorithm in the "Model Building" sub section 3.2.2. Algorithm: ECA_ResNet-based Maize Pest Identification Model
The model provides a clear and concise outline of the steps involved in constructing the ECA-Adam-ResNet50 model.
- What do you mean by "4.1. Subsection"?
R.: Sorry for our lapse, this was due to formatting within the MDPI model.
- How the numerical results are calculated and evaluated in Table 1.
R.: The main idea of ablation experiments on ResNet50 models with different improvement methods under the same dataset is to control the variables. To prove that both the fusion channel attention mechanism and the Adam optimizer are meaningful, respectively, the following method is used for validation:1. Add the ECA attention mechanism on top of the ResNet50 model to analyze the accuracy effect. 2. Add Adam optimizer on top of the ResNet50 model to analyze the effect of the accuracy rate. 3. Add both the ECA attention mechanism and Adam optimizer on top of the ResNet50 model to analyze the effect of the accuracy rate.
- Every equation should end with "," or ".".
R.: Done, thank you for your improvement.
Round 2
Reviewer 1 Report
The manuscript has been improved
Reviewer 2 Report
In line 201, delete Equation. Put only (1)
In line 232, delete Equation. Put only (2)
In line 238, delete Equation. Put only (3)
In line 244, delete Equation. Put only (4)
Minor
