Recognition of Strawberry Powdery Mildew in Complex Backgrounds: A Comparative Study of Deep Learning Models

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

1. I consider it necessary to provide information about the performance of the models in terms of TeraFLOPs.
2. How feasible is the implementation of the models in real agricultural tasks? Is highly specialized equipment required? I believe it would be a valuable contribution to the work to provide an overview of their field implementation, beyond controlled experiments

Author Response

请参阅附件。

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Here are the things that I like about this paper:

• The authors have done a great job in tackling a real-world agricultural problem by leveraging advanced technology. The use of 14 different CNN models to evaluate their performance in complex field conditions is thorough and provides a comprehensive understanding of each model's capabilities.
• The paper is well-structured, and the methodology is clearly explained, making it easy to follow.
• The results are promising, especially with models like InceptionV4 and DenseNet-121 showing high accuracy and recall rates.
• The inclusion of HSV-based image segmentation to enhance model performance is a smart move, and the discussion on inference time provides practical insights for real-world applications.

Areas for improvement

• The paper could benefit from a more detailed discussion on the limitations of the current models and potential solutions. For instance, while SqueezeNet's poor performance is noted, exploring why it underperformed and how it might be improved would add depth.
• The paper could include more on the practical implications of deploying these models in real-world settings, such as cost considerations or integration with existing agricultural practices.
• It would also be helpful to see a comparison with traditional methods of disease detection to highlight the advantages of the proposed approach.
• While the paper mentions future work, a more detailed roadmap or specific next steps would provide clearer direction for ongoing research.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The paper focused on the identification and prediction needs of strawberry powdery mildew, and conducted comparative identification research on strawberry powdery mildew under complex backgrounds. The paper established a strawberry image dataset and used HSV segmentation method to segment strawberry images, improving recognition performance. The effects of 14 deep network recognition methods were compared. The paper has certain application prospects in the identification of strawberry powdery mildew.

There are still some issues with the paper:

1. Although the paper establishes a dataset and compares the recognition performance of different deep learning methods, overall it adopts existing algorithms and innovation is not significant.
2. There is a lack of example images of powdery mildew, the paper does not provide comparative images of healthy strawberries and powdery mildew strawberries, lacking intuitive perception. And the paper states that a long period of health image collection was conducted. What is the representativeness of these images? What are the characteristics of strawberry leaves at different growth stages? The paper explains that images of powdery mildew are collected through the internet, so how are healthy images and images with powdery mildew related and integrated for utilization? What are the characteristics of powdery mildew in different periods or infection periods, and how are the images labeled? More example images and annotated images should be provided, such as images of different growth stages and images of powdery mildew at different infection stages, to illustrate the differential characteristics of images at different stages.
3. The paper description: "The incubation period of powdery mildew is not a violation." So, what are the image features of powdery mildew in the early stages of infection, and what is the difficulty of its recognition? Is it possible to make predictions through images?
4. The paper only provides data on the recognition accuracy and running time of different recognition methods, lacking a more in-depth and detailed analysis. For example, how effective are different methods in identifying powdery mildew at different infection stages? It is best to provide recognition examples for each algorithm and compare and analyze the recognition effects of different algorithms on healthy and infected leaves, in order to provide readers with a more intuitive judgment. It is best to conduct comparative analysis of the effects of different algorithms and modules in order to select appropriate algorithms and modules for powdery mildew recognition.
5. Can the HSV based image segmentation method distinguish green objects such as broad-leaved weeds?
6. In Figure 6, the same symbols cannot distinguish different methods. It is recommended to use a combination of line type, color, and symbol to ensure that the symbol for each method is unique. Figure 7 is not clear enough.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Comments and Suggestions for Authors

The questions that have been raised have received appropriate responses, and there are no other issues.