Detection and Characterization of Stressed Sweet Cherry Tissues Using Machine Learning

Round 1

Reviewer 1 Report

This paper has a very big title. However, the workload of the paper does not match the title.

 

The whole structure of the paper is bad organized. The authors talked a lot about the machine learning development at the very beginning of introduction. However, they should first place the study purpose (Detection and Characterization of Stressed Sweet Cherry Tissues), and why it is important; Then introduced the current state of the research (what algorithms were used to solve the problem) and the innovation of the authors’ own algorithm. You should have a practical problem first and try to find a proper method to solve it. 

 

In addition, “2 Contributions” should be included into Introduction part. “3.1“should go to introduction part as well. “3.2” should go to materials and methods part.

 

I don’t see the need of last paragraph Line 92-97 in the introduction.

 

The acronym ERICA at Line 102 must be explained when it first appears in the text.

 

How the images were acquired? The authors didn’t provide necessary description. How did the authors locate the leaves and trees with high precision? Did they use RTK? It looks to me that the images were not collected by drones if it is submitted to “Drones”.

 

The abstract has limited useful information to readers. The reviewer suggests that the authors rewrite the abstract after reorganizing the whole structure of the paper. No reference should apper in the abstract.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The manuscript entitled “Machine Learning Algorithms in Perennial Fruit Trees: Detection and Characterization of Stressed Sweet Cherry Tissues using Precision Agriculture” still needs some improvements to meet the required quality to be published in the Drones journal. First, the manuscript requires moderate English language editing. Besides, the description of the methods must be improved, with a more detailed explanation of each step. Moreover, some confusion between methods and results is present in the manuscript. Improve also the quality of the discussion with references to results obtained by other authors. Finally, and into more detail, there are other issues that need to be addressed, such as: 

• In section 2 there is no description on what kind of images belong to ERICA, nor on how were they acquired;
• Line 120: you mention three optical factors, but I believe that you mean four!
• Lines 125-126: strange sentence!
• Figure 4: I suggest a unique graph with different colors to distinguish between training and test datasets. Also, improve the graphical quality of the image!
• Line 171: reference to Figure 5 before the reference to Figure 4 doesn’t make sense!
• Lines 221-222: explain the purpose of the training, validation, and test datasets, here mentioned.
• Lines 230/231: these 24 experiments should have been mentioned in the Methods section!
• In Tables 1 and 2 too many decimal digits (not significant!) are considered for the accuracy metrics.
• Lines 237/244: The metrics used to evaluate the predictions should also have been mentioned in the Methods section!
• Section 6 (Ablation) should be a subsection of section 5!
• Line 208: 27 or 24 models?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The author has carried out a very practical study, produced the ERICA dataset, and used deep learning algorithms to achieve good results in the detection of stressed sweet cherry tissues, but there are still current problems in the manuscript:

1. Figure 3 does not explain the structure of resnet50 well. It is recommended to increase the details of the model.
2. Improve the clarity of Figure 4 and unify the font format in the image
3. Please explain why the author did not modify the yolov5 and Resnet models based on his own dataset, because modifying the yolov5 and Resnet models based on the existing model based on past experience will improve the recognition accuracy.
4. The author, please explain why the yolov5 and Resnet models were chosen instead of other deep learning models.
5. How does the author distinguish that the leaves in the dataset are infected? Is it based on the shape and color of the leaves or other characteristics? Please explain.
6. Please explain how the data set is made.
7. How are the images of the dataset obtained? Do you use drone aerial photography or camera shooting or interception from the video? Please explain.
8. The 230 lines of the manuscript contain 24 models, and the 298 lines contain 27 models. Please check the description.
9. Please indicate the geographic location of data collection in the ERICA data set.
10. Is the ERICA dataset produced by the author universal? That is to say, it can be applied to the detection of stressed sweet cherry tissues in other regions.
11. As a research paper, the contents of Part 4 Materials and Methods and Part 5 Results and Discussion are less, so it is recommended to supplement them.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Thanks a lot for your revision. It is not enough though.

This paper has a very big title by saying "precision agriculture". However, the workload of the paper does not match the title. Please change.

The structure of the pape still needs further improvement. Figure 1 and Figure 2 and the related description belongs to the Methods and Materials part. The background belongs to the Introduction.

The images were not collected by drones. How is this study related to drones since it is submitted to “Drones”?

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Some problems with the manuscript´s structure remain. Text in lines 96-125 (including actual Figure 1 and Figure 2, former Figure 5 and 4, respectively) is too detailed for the introduction section. The workflow of the methodology (Figure 1) should be the first thing after the beginning of actual section 3! In the previous version of the manuscript, the description of ERICA was well placed (although not explained in detail!), there was only the need to introduce the acronym and explain it briefly in the introduction’s section! So, subsections of section 3 should follow the sequence in Figure 1, starting with subsection 3.1 with the description of the test area (include a figure with its location!) and a detailed description on how ERICA data was acquired and pre-processed to be fed into the classification algorithms. Still, some sentences require English language editing. For example, the sentence in lines 103-104 should be written as follows: “To obtain the samples from the trees there is the need to follow a strict process.” After these modifications, I believe that the manuscript has enough quality to be published.

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The author modified some of the previous comments and met the requirements, but there are still some questions that the author's answers are not satisfactory. I think the author’s answers did not solve the problem very well. The parts that still need to be modified:

1. Figure 5 is still too simple to reflect the network structure of ResNet50. It is recommended to replace the structure diagram and add a text description.
2. The manuscript did not reflect the author’s attempts to change various hyperparameters but the results did not change significantly.
3. As to why yolov5 and Resnet models were chosen, the author explained that it is because Yolov5 is a new machine learning algorithm, which shows that it is not convincing. Choosing yolov5 and Resnet models should be considered from the accuracy and performance of the two.
4. The author explained that special equipment is used to obtain images, please provide the information on the image acquisition equipment
5. Can the geographic location acquired by the ERICA dataset be better displayed to readers in the form of a map? Ask the author to consider.
6. Is there any relevant research to prove the author's "the symptoms of Armillaria are independent of the location of the field"?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf