Real-Time Counting and Height Measurement of Nursery Seedlings Based on Ghostnet–YoloV4 Network and Binocular Vision Technology

Round 1

Reviewer 1 Report

Dear Authors,

I have finished my review on the paper. I find the topic to be of most relevance for the international audience and with some work the paper could be a good contribution. However, there are several important issues which need to be addressed both in phrasing and reporting as well as in language used. In addition, there are needed substantial format corrections and improvements. Parts of the sections are quite mixed in this version, lacking coherence in reporting. The reader may find a mix of materials and methods with results in the results section. Materials and methods are too wordy and less informative and structured. Introduction fails largely to focus the reader on the topic and to understand its importance by adding rather unnecessary applications of deep learning. Discussion seems to follow the same trend by not addressing important issues at least for the practice. In my opinion, the paper needs substantial writing and configuration before acceptance. Please check my supplementary point-by-point comments given in the following and good luck!

Best regards,

R.

 

Title:

Seems to be OK.

Abstract:

Second line: I am unsure if manual measurements could be described as inaccurate. Also, this is in contradiction with the accuracy of the proposed framework which accounted for 92%.

Introduction:

Second paragraph, first line: I am unsure if seedling production is an important part of the precision forestry and not vice-versa, the ‘precision forestry’ could be an important approach to seedling production counting/assessment.

Second paragraph at all: lacks important reference data by citation since the Authors argue on inefficiency of the manual approach to the problem. Each sentence should be provided with references by citation to make a compelling case.

Third paragraph: uncarefully written and rather confusing. Why this enumeration of DL approaches in several topics and not focusing exactly the reader on the state-of-art in nursery production and seedling to guide the taught to the lack of relevant studies in the field? Please check for instance these, as well as other many helpful resources:

https://www.mdpi.com/1999-4907/12/6/782

https://www.mdpi.com/1999-4907/9/11/665

Fourth paragraph, first sentence: it depends on what ‘higher accuracy’ means. For instance, an accuracy of 85% may leave out 15% in counting plants. I am not sure if the nursery managers would like that...

Fifth paragraph, second sentence: are there any studies to support the statement?

Fifth paragraph, at all: looks more as conclusions instead of problem definition and goal formulation.

Materials and methods:

First statements: it is uncommon to put information without sub-chaptering. Maybe this, along with some more explanations would qualify as text and figures for a new subsection on the architecture of study.

2.1.1. Data collection:

Latin names of species are missing. Not sure what large, medium and small stand for. Needs clarification. The description of sampling procedure is confusing. Figure 2: I am not sure what a ‘sprucer’ is.

 

2.1.2. Data enhancement

I am not sure if ‘data enhancement’ is the right concept. Please check. Is it better to use ‘data augmentation”? Table 1 – is not a table as it shows pictures. Better to convert it in a figure. Yet again, the description of the process is confusing. You have taken 5 by 500 images resulting in 2500 images as of point 2.1.1, then you have augmented the images by 15 procedures, that is 2500 by 15 = 37500 images of which 2500 were used for validation and 22500 for training. However, 2500 + 22500 = 25000. Language used in this section is confusing. For instance: ‘The effect test...’

2.2. Experimental configuration:

Is it better ‘experimental design’ or ‘experimental architecture’? Reading the text, it seems to be the hardware and software architecture…

2.3. seems to be a general description in most of its parts and needs citation.

2.3.1. Some statements need citation. For instance, that saying that Ghostnet outperforms other architectures. Figure 3 and text describing it: needs citation. Part of the text following Figure 3 is redundant.

2.3.2. First two paragraphs are confusing and the language used does not indicate if these were methods or is something that is known. The Authors should improve these parts by citation or by using the proper tense. Same apply to the following paragraphs from this section.

2.3.3. It is unusual to start a new section by a figure. Text is confusing.

2.4.1.

Definition the sapling height is confusing and uses improper terms. I am not sure if Figures 5 and 6 are necessary. However, ‘d’ is not shown in figure 6.

2.4.3.

Needs substantial rephrasing to sound like a paper.

Results:

Uncommon to start a chapter without a subsection. Anyway, most of the text from here is rather methods and not results. Key performance indicators used to evaluate the accuracy were not defined in the materials and methods which were rather wordy and focused on rather unessential things.

3.1. Visualization

First sentence is redundant. Second and third sentence contain descriptions which are materials and methods. Fifth sentence is poorly written. Rest of the sentences are poorly written to understand if they are actually comments of the results or methods used. Images given are of poor quality.

3.2. The text is a mix of results and materials and methods.

3.3. Same as above

3.4. Same as above

Discussion:

Unusually written. The sections could be merged and the language needs checking. Also, other external factors which are specific to nurseries are not addressed. How about very short seedlings which are overrun by weed? Can they be detected? How about the height measurement errors and their contribution to nursery management? Are these important? In my opinion, the nursery management targets the size of the seedlings to remove them from nursery… Also, there is the question on how well the species could be identified…Then, is the accuracy of 92% enough to support precision forestry in nurseries and nursery management? If yes, how? Based on the results, when should be the measurements done? Also, what would be the applicability of the system to larger areas and how many resources would be needed?

Conclusions:

This section seems to be quite long.

Author Response

lease see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

Real-time counting and height measurement of nursery seedlings based on Ghostnet-YoloV4 network and binocular vision technology

In this work, authors combine binocular stereo vision with Yolo in order to allow analsis of palnt heights.

Authors use ZED: they should mention which performances they expect using ZED instead of ZED2 (see e.g.: DOI: 10.3390/horticulturae8010021)

Authors use Yolo v4: they should mention which performances they expect using V4 instead of V5 or other (see e.g.: DOI:10.3390/agronomy12020319)

Please explain what is and how it works the BM algorithm for stereo matching. 

Authors implement data augmentation strategy in order to verify the performance of the method. Since the two image of the stereopair are captured at the same time, the authors should modify images through a coupled strategy (i.e. if one image is tilted, also the other one should be tilted, with some change due to the different position): this is not clear from the paper. Please explain in amore clear way. 

In case of field with highly dense plants, stereo reconstruction might fail. Authors should better investigate the case of areas with a high density if plants, even with partially occluded portion of images (e.g. plants staying behind other plants).

It is not clear how the basis of the plants can be defined in case of high and highly dense grass at the basis of seedlings.

The manually measured height TH is different from thee height H measured by the system. Authors should better explain when substantial differences might be found and how to compensate them.

What diffeerent performances can be expected in case of biggere/smaller plants or in the case of different varieties. 

More in genereal an uncertainty analysis is missing. 

 

 

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The paper has been improved but still there are some points to be addressed:

References are badly linked: in my version I can read "[Error! Reference source not found.]"

REFEREE: Authors use ZED: they should mention which performances they expect using ZED instead of ZED2.

AUTHORS: Response 1: Check the paper again, it has been revised accordingly.

REFEREE: I know there was already some line on that, but I think the authors should be more accurate and give some quantitative data. 

 

REFEREE: Please explain what is and how it works the BM algorithm for stereo matching.

AUTHORS: Response 3: A description of the BM algorithm has been added accordingly.

REFEREE: I am still missing a technical description of the BM algorithm.

 

Please highlight with yellow colour the revised text

Author Response

1. All citation errors with "[Error! Reference source not found. Fixed

2. Added a paragraph about the article to the article, clearly explaining the experimental route and the goals, effects and errors achieved by the experiment.

3. A description of the specific workflow of the BM algorithm has been added to the text in subsection 2.5.3.

Author Response File: Author Response.docx