Revolutionizing Coffee Farming: A Mobile App with GPS-Enabled Reporting for Rapid and Accurate On-Site Detection of Coffee Leaf Diseases Using Integrated Deep Learning

Round 1

Reviewer 1 Report

The manuscript presents a solution to assist coffee growers that is based on the use of a mobile/web application that is able to detect a series of coffee diseases with the help of models based on deep learning such as Resnet50, DenseNet and others.

In the introduction, it is stated that the goal is to support farmers (line 113), and then the contributions are presented (probably contribution 1 is to assist farmers, please check spelling).

However, in section 3. Material and Methods, the authors present and discuss the data gathering from farmers, but details or statistics extracted from the survey seem to be of little interest to the problem of coffee disease detection (i.e. can the data collected from the survey be used to improve the deep learning classifier performance?). Furthermore, it would have been welcome to present some images of several classes.


Regarding the deep learning models used, they are not the latest state of the art (if their corresponding peers were cited properly, the year would be before 2023). And the pipeline to deploy the models on mobile/cloud has limited specific information, missing details about the implementation.

The English language should be properly checked for errors.

Author Response

Dear Reviewer, 

Thanks for your valuable comments. They contributed a lot to enhance the article.
Attached are the details.

Regards.

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors introduce an application for on-site coffee leaf infection detection, featuring GPS reporting capabilities. This implementation leverages transfer learning models and Docker for deployment at the edge. The paper is well structured and presented. Notably, the results demonstrate a significant improvement over the current state-of-the-art methods. Such kinds of contributions should always be valued and welcomed as they can be of great help to upcoming research efforts and benefit in the general research community.

However certain improvements can be made: 

-> Dataset description, metrics:

Is there a link to the dataset?

What distribution does the dataset have in terms of infected/non-infected images?

What kind of preprocessing was made?

F1 is probably a better metric to compare the models.

-> Comparison to related work, uniqueness of the approach: Point out more clearly the difference to other approaches and the contribution of this paper.

Author Response

Dear Reviewer, 

Thanks for your valuable comments. They contributed a lot to enhance the article.
Attached are the details.

Regards.

Author Response File: Author Response.pdf

Reviewer 3 Report

This study introduces a pioneering mobile application equipped with global positioning system (GPS)-enabled reporting capabilities for on-site coffee leaf disease detection.However, there are still several problems that need to be improved.

1.The ablation experiments carried out in this paper are not sufficient.

2.However,the paper is a little brief in the future-looking section and does not delve into the possible challenges of this technology or the possible future directions.

3.In addition, the article will be more persuasive if it can provide some practical application examples or feedback from farmers.

4.Besides, although the performance of the model has been verified by experiments, the selection basis and optimization process of the model are not discussed in depth. The selection of a deep learning model usually involves trade-offs in terms of model complexity, computing resource requirements, training time, and other aspects. Therefore, it would be possible to further explore the rationale for model selection and possible optimization methods in the article, which would make the article more in-depth and breadth.

This study introduces a pioneering mobile application equipped with global positioning system (GPS)-enabled reporting capabilities for on-site coffee leaf disease detection.However, there are still several problems that need to be improved.

1.The ablation experiments carried out in this paper are not sufficient.

2.However,the paper is a little brief in the future-looking section and does not delve into the possible challenges of this technology or the possible future directions.

3.In addition, the article will be more persuasive if it can provide some practical application examples or feedback from farmers.

4.Besides, although the performance of the model has been verified by experiments, the selection basis and optimization process of the model are not discussed in depth. The selection of a deep learning model usually involves trade-offs in terms of model complexity, computing resource requirements, training time, and other aspects. Therefore, it would be possible to further explore the rationale for model selection and possible optimization methods in the article, which would make the article more in-depth and breadth.

Author Response

Dear Reviewer, 

Thanks for your valuable comments. They contributed a lot to enhance the article.
Attached are the details.

Regards.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Thanks to the authors for considering the suggestions and improving the manuscript accordingly - it is an improved version of the manuscript overall. My observation that the artificial vision models used were not SOTA was not clear, and it would have been useful to experiment with recent architectures, such as those based on Visual Transformers.

It is necessary to include an experiment with a more recent architecture, such as Visual Transformer, and compare and analyze the results.

The English language can still be improved.

Author Response

Attached is the reply to the feedback shared.

Author Response File: Author Response.pdf