An Automated Framework for Plant Detection Based on Deep Simulated Learning from Drone Imagery

Round 1

Reviewer 1 Report

The authors improved their manuscript extensively; however, still more improvement requires before I can recommend this manuscript for publication.  

Ln 76- What is the primary purpose of counting plants? More elaboration is required.

Ln 187: Training-Patch Generation is not straightforward! The authors should provide their method with details. 

Ln  272: Drone Imagery. Authors should provide all image acquisition details, like exposure, ISO, etc. If image acquisition did automatically, try to extract information from the Exif file!

Section 2.7 is not clear to me. Does authors applied the developed methods in literature on their dataset, or compare their results with published results. If it is a later one, I recommend to remove the subheading ( Comparative Study) and keep the text as the main part of the discussion.

Author Response

Please see the answer's file.

Author Response File: Author Response.pdf

Reviewer 2 Report

Dear Authors: Thank you for your revision and the answer letter. It is confirmed as it is. Best,

Author Response

Thank you. We appreciate the time and effort that you have dedicated to our manuscript.

Reviewer 3 Report

Now the manuscript has substantially improved with the modifications made by the authors.

Author Response

Thank you. We appreciate the time and effort that you have dedicated to our manuscript.

Reviewer 4 Report

Dear author and editor,

The manuscript proposes a methodology for imaging-based plant detection using self-training tools. The methodology, developed for corn, has the potential to be implemented for other crops.

The paper is well organised and clear, although English writing must be reviewed. The topic meets the Journal scopes.

Here are some specific comments:

Line 16: “deep learning” should be DL

Line 18: why do you state remote sensing provide ground-truth data? It’s a contradiction.

Lines 25-30: this is a repetition of the previous paragraph

Lines 58-59: this is a repetition of what stated in lines 55-56

Lines 66-68: could you provide some example of this drawbacks?

Line 76: remove “a” before “few”.

Line 96: define R-CNN

Line 127-132: check concordance of verb-tense

Lines159-161: could you please rephrase this sentence? It’s not clear to me what you did to overcome the issue.

Lines 176-177: “which are appeared” should be “which appeared”. Also, I suggest standardising verbs (either past or present tense all across the manuscript)

Line 275: what do you mean by “regular format”? Moreover, could you provide information on the phenological stage? This would be important to understand the value and solidity of methodology. As you stated in the Discussion section, the issue of this method could be that it depends on the colour of the plant (and the plant's size). Therefore, if you provide us with this information, this would be very useful.

Line 286: it would be interesting to understand why you chose this k value. You were mentioning above that this choice is difficult, and you were explaining the meaning of choosing 0.5. Why don’t you try to explain what changes if instead of 0.5 you use 0.2?  

Line 419: I guess it also depends on the size of the plants

Author Response

Please see the answer's file.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors addressed all raised issues, therefore I am recommending this paper to be accepted for publication at the current form. 

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

The abstract needed enhancement, I recommend added a section regarding the results in the abstract

Line 77-82 and 87-89 ,91-93 ,107-111, 267-274 the Font size needed to unify

Line 342 to 353  I recommended to move up the material and methods section

Figure 3 needed enhancement the resolution of numbers in  x and y lines

Figure 5 needed to enhance the resolution text

Reviewer 2 Report

The manuscript claims to develop a framework for automatically generate the training set with the label for plant detection using deep learning. The presented method is limited to green plants, also as authors mentioned, the Hough transform (HT) is not reliable and might require manual processing. 

The manuscript is written very poorly. The introduction is long and disorganized. The framework didn't clearly explain. The experimental part also not described clearly; for instance, What plant authors used in their study? Does the plant's species impact the algorithm? Does the algorithm need to be modified if the plant's changed? The "Comparative study" requires more elaboration. The conclusion is the reputation of the abstract and results. The authors need to rewrite it. 

Lnn 53: It might be applicable to use UAV in cloudy weather where active sensors are attached, but I am not sure it provides useful data with passive sensors!

Ln 136: How it improved? The authors didn't provide any supporting background in their manuscript?!

Figure1: The Deep model framework is not clear; I am suggesting to improve your figure to present more information. 

Does the presented algorithm work only on small plants?

Ln 172: What are R, G, B? provide the full term together with its abbreviation in the first place you are using it!

Ln 212: Citations in the text need to be checked. 

Figure 5 and 1 can be merged. 

Ln 288: remove the link and provide the appropriate citation. 

Reviewer 3 Report

Dear Authors,

First of all, I must mention that your manuscript attracted my attention. All the research steps and results have been designed well enough, although there are some minor points that I have mention in the following:

1- Figure 1 is too simple for such deep learning job and also, publishing in 2020 and you have to enhance it regard to the year that you want to publish.

2- There are too many formatting errors like references in pages 6 and 15, empty space like pages 7 and 17, several font styles and sizes in page 19, and etc. Therefore, you have to either revise the manuscript carefully before submitting and publishing or editing the manuscripts by some journal's section expert.

Kind Regards,

Reviewer 4 Report

An Automated Framework for Plant Detection based 2
on Deep Simulated Learning from Drone Imagery

In this paper an automated framework for detecting and counting plants in the RGB 457 images captured by drones have been proposed. In this framework, the training patches were made and extracted using 458 automatically extracted single plants based on ExG and using C-means clustering. Also, the HT 459 algorithm was used for cultivation line detection to prepare the image scene for plant detection. The 460 core of this framework is Faster-RCNN, which is trained by simulated data in order to detect plants 461 from drone images.

Generally this work is well done. In the abstract (lines 26-30) and in the introduction (lines 136-143) the Authors could better highlight which is the novelty in this paper.

Introduction

Lines 41, 42: In this context, the Authors can also consult the following paper:

Servadio, P.; Verotti, M. Fuzzy clustering algorithm to identify the effects of some soil parameters on mechanical aspects of soil and wheat yield. Spanish Journal of Agricultural Research 16 (4), e0206,  (2018) eISSN: 2171-9292. 
https://doi.org/10.5424/sjar/2018164-13071

Instituto Nacional de Investigación y Tecnología Agraria y Alimentaria, O.A, M.P. (INIA)

Authors see line 212 and line 394:

Line 212: Error! Reference source not found. ????

Line 394: Figure 16Error! Reference source not found ???