Next Article in Journal
Assessing the Role of Crop Rotation in Shaping Foliage Characteristics and Leaf Gas Exchange Parameters for Winter Wheat
Next Article in Special Issue
Hardware and Software Support for Insect Pest Management
Previous Article in Journal
The Use of Temperature Based Indices for Estimation of Fruit Production Conditions and Risks in Temperate Climates
Previous Article in Special Issue
Predicting the Occurrence and Risk Damage Caused by the Two-Spotted Spider Mite Tetranychus urticae (Koch) in Dry Beans (Phaseolus vulgaris L.) Combining Rate and Heat Summation Models for Digital Decisions Support
 
 
Article
Peer-Review Record

EfficientDet-4 Deep Neural Network-Based Remote Monitoring of Codling Moth Population for Early Damage Detection in Apple Orchard

Agriculture 2023, 13(5), 961; https://doi.org/10.3390/agriculture13050961
by Dana Čirjak 1,*, Ivan Aleksi 2, Darija Lemic 1 and Ivana Pajač Živković 1
Reviewer 1:
Reviewer 2: Anonymous
Agriculture 2023, 13(5), 961; https://doi.org/10.3390/agriculture13050961
Submission received: 7 March 2023 / Revised: 4 April 2023 / Accepted: 24 April 2023 / Published: 26 April 2023
(This article belongs to the Special Issue Hardware and Software Support for Insect Pest Management)

Round 1

Reviewer 1 Report

Smart System for Coding Moth Monitoring Based on Artificial Neural Networks

 

This work introduces a smart system designed for monitoring coding moths in apple orchards, which consists of a portable trap and a computing device equipped with an artificial neural network, which can periodically transmit the result of the occurrence of moths. The system performs well on the dataset provided in the work and has been applied in the natural environment.

 

Although various fixed field pest monitoring systems have been widely applied in many places and the workflow of these systems is similar to this work, the designs of this work, including portability, front-end calculation, and energy-saving, provide new ideas for the promotion of more application scenarios. However, to better demonstrate the effectiveness of the system, some parts still need to be improved and corrected. Specific comments are as follows.

 

1.      The title “based on artificial neural networks” is too broad. The authors should read related applied research and summarize the unique highlights of the work so that the title can express the core work of the paper.

2.      The concept of “artificial neural networks” is enormous. Considering that the work only uses EfficientDet, it is recommended to use the term “deep neural network” instead.

3.      In Table 1, it is not corrected in the dataset allocation ratio. In general, the test set should be larger than the validation set. The test set should reach a ratio of about 1:4 with the training set, and the validation set should be much smaller than the test set. Because it is necessary to test the overall performance of the model through a large enough dataset, while the purpose of the validation set is only to observe whether the model is overfitting. Too large a validation set will cause long validation time after each epoch, affecting the entire training time. Therefore, the authors should re-experiment based on the newly split test datasets.

4.      In 258-260, the author should give an experimental explanation of the reason for choosing EfficientDet.

 

5.      In 340-342, the author should give evidence or comparison to verify the “consumes less energy” conclusion.

Author Response

Dear Reviewer 1,

I would like to thank you for your time and effort in reading our article. Also, for your comments and suggestions.

I have accepted all your suggestions and incorporated them into the manuscript.

Sincerely,

 

Dana Čirjak, M.sc.

University of Zagreb, Faculty of Agriculture

Department for Agricultural Zoology

Svetošimunska cesta 25

Zagreb 10 000

Croatia

Author Response File: Author Response.docx

Reviewer 2 Report

The paper focus on an automatic system for codling moth monitoring, in particular on the process for automatic recognition of the moth into the trap.

In general, the manuscript is clear, the English language linear and understandable also when informatic models are explained, the methods sufficiently described and the results reported and commented in extensive way. The paper focuses greatly on the automatic recognition and is lacking in the hardware functioning of the system. For example, no tests were presented about the power consumption under different conditions, even if Authors stated that the transmission of images is power and time consuming.

In the Introduction, I suggest Authors to better define the purpose of the monitoring using traps, for example to identify the time the flight starts in the field or the trend of catches or the definition of empirical economic thresholds to apply insecticides.

Considering that several papers deal with automatic monitoring of codling moth, at the end of the Introduction, Authors should explain precisely what is the novelty of this paper and the contribution that they intend to give in this specific case. This, also respect the methods used in the paper [56] from the same Authors, that look the same.

In Materials & Methods, define Average Recall and Average Precision in the text. The same for IoU in Table 2.

Did you have other moth species in traps similar to codling moths? If so, in defining the labels, did you consider this possibility?

Author Response

Dear Reviewer 2,

I would like to thank you for your time and effort in reading our article. Also, for your comments and suggestions.

I have accepted all your suggestions and incorporated them into the manuscript.

Sincerely,

 

Dana Čirjak, M.sc.

University of Zagreb, Faculty of Agriculture

Department for Agricultural Zoology

Svetošimunska cesta 25

Zagreb 10 000

Croatia

Author Response File: Author Response.docx

Reviewer 3 Report

Dear Authors,
You have presented an interesting paper using artificial neural networks to analyse image-photos of insects occurring in agrocenoses.
The paper is interestingly written and contains a very interesting extended introduction. In my opinion, some basic information was missing: characterise the type of neural network chosen. Provide information defining the quality of the network.
Please prepare a proper discussion of the results. In its present state it is unacceptable. Please cite at least 15 additional scientific sources on similar topics. Please paste a map indicating the experimental location.

Author Response

Dear Reviewer 3,

I would like to thank you for your time and effort in reading our article. Also, for your comments and suggestions.

I have accepted all your suggestions and incorporated them into the manuscript.

Sincerely,

 

Dana Čirjak, M.sc.

University of Zagreb, Faculty of Agriculture

Department for Agricultural Zoology

Svetošimunska cesta 25

Zagreb 10 000

Croatia

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

In the revised version, the authors provided a detailed response to the previous comments and revised them carefully.

There is a small error that the author needs to modify. In the confusion matrix shown in Figure 10, the color change should correspond to the accuracy rate (percentage) rather than the quantity.

 

Author Response

Dear Reviewer 1,

I would like to thank you for your time and effort in reading our article. Also, for your comments and suggestions.

Sincerely,

 

Dana Čirjak, M.sc.

University of Zagreb, Faculty of Agriculture

Department for Agricultural Zoology

Svetošimunska cesta 25

Zagreb 10 000

Croatia

Reviewer 3 Report

I accept

Author Response

Dear Reviewer 3,

I would like to thank you for your time and effort in reading our article. 

 

Sincerely,

 

Dana Čirjak, M.sc.

University of Zagreb, Faculty of Agriculture

Department for Agricultural Zoology

Svetošimunska cesta 25

Zagreb 10 000

Croatia

Back to TopTop