Apple Shape Detection Based on Geometric and Radiometric Features Using a LiDAR Laser Scanner

Round 1

Reviewer 1 Report

Apple shape detection based on geometric and 2 radiometric features using a LiDAR laser scanner

Major Revisions due to plagiarism (Rating out of 100 (60=best)) 43(=71%)

(1-10(best))
Novelty: 7
Writing&Language: 7
Scientific Rigor: 9
Introduction&Literature: 6
Method&Results: 8
Discussion&Conclusion: 6
Deductions: For plagiarism, I have to give it a major revision, as this is scientifically not acceptable, but can be made acceptable.

In short: Your paper proposes a method to detect apples on trees, using LIDAR.

Expertise of Reviewer: I am not currently performing research in the field of remote sensing; however, I lectured it until recently. I am currently developer and researcher in photogrammetry and LIDAR.

Overall Statement:
Plagiarism Scan with Plagscan returned some plagiarism.
You are not citing correctly. Just mixing the word order and putting a reference behind is not the correct way of citing others people work. This is clear plagiarism! Please check and correct that.
Example:
You: "An inertial measurement unit (IMU) (MTi-G-710, XSENS, Enschede, the Netherlands) was placed 0.24 m aside from the LiDAR to monitor the 3D tilt of the tractor."
"The three-dimensional tilt of the tractor was monitored by an inertial measurement unit (IMU) MTi-G-710 (XSENS, Enschede, Netherlands), which was placed 0.24 m aside from the LiDAR"
(in addition, this whole paragraph is plagiarism)

Very well described scanning setup with a thorough analysis of the measured values and clear representation of the results. However, i struggled strongly with the huge amount of abbreviations and would like to ask you to reconsider if certain abbreviations should be re-mentioned in the text in their spelled out form. Thus allowing the reader to read on instead of permanently looking backwards what each abbreviation stands for. Even if they are common in botanic this journal is for remote sensing so not all will know the abbreviations by heart.
Furthermore, your Conclusion is not a conclusion but an abstract. A conclusion should state whether the aim set out to achieve was achieved or not. Your aims and objectives is:
"The aim of this study is to propose a methodology for segmenting, localizing, and analyzing fruits on the tree, based on the assumption that apples show enhanced RToF at 905 nm compared to foliage and woody parts. The main objectives are: (1) the classification of 3D points of apples based on reflectivity and geometric features derived from LiDAR laser scanner; (2) the development of a fruit segmentation algorithm that allows measuring the fruit size in defoliated trees; and (3) the evaluation of the proposed technique on foliated apple trees."
So, did you achieve to create a method for segmenting, localizing and analysing fruits on the tree, were able to classify them and so on..?

Detail Feedback:
Line 103: "(Latitude: 52.466274° N, Longitude: 12.57291° E)" Do you really think that six digits are necessary?! Furthermore, checking on the location as provided by you I end up in lake Beetzsee in both, Google Maps and OpenSteetmap, while I found your institute at N52.44, E13.01 in both map services.

Line 159: Why did you normalise between 0-100 and not like normal between 0-1?

Figure 1 and 2 appear in low pixelated quality. All other figures seem to be of sufficient quality. Please check. In general, the font type used in figures should be the same as in the publication text and of the same size. Please have a look especially as your text is often smaller in the figures and could be increased without problems.

Suggestion for Figure 1 and 2 to point out the exact differences in the figure, e.g. by different shading of the different cells?

Line 199: "variables,, Dd[...]" typo.
Line 241: "date.The" typo
Figure 6. If possible, it would be great if you could change the lines of the plot according to the lines in Figure 5, for people that print in b/w that they can see straight away, which belong together.

Author Response

Please see the attachement, since figures cannot be uploaded here.

Author Response File: Author Response.docx

Reviewer 2 Report

In this work, a LIDAR system is used to scan the apples before and after defoliation. An apple detection and location methodology based on reflectance intensity is proposed, as well as geometric parameters such as linearity and curvature obtained from the point cloud. The perimeters of the fruits are also measured obtaining a good adjustment with manual measurements.

The methods used are adequate and the subject is currently of great interest. In my opinion, the main contribution of this work is the application of LiDAR sensors to estimate the fruit size. However, to consider publishing this work, the following issues should be addressed.

Major Comments

• Lines 303-304. “In the upper sections, ΔW3-2 and ΔW4-3 of the canopies where the fruits grow more distinctly, the analysis of apple size resulted in R2adj = 0.95 with a RMSE = 3.2 %.” It can be seen in Table 1 that previous results correspond to two particular cases (DAFB₁₀₄ / ΔW_4-3 - R2adj = 0.95) and (DAFB42 / ΔW4-3 - RMSE = 3.2 %). In the remaining cases, the results are lower in the range of R2adj = 0.69-0.95 and RMSE = 3.2-7.7%. Therefore, this sentence should be rewritten to clarify this point.

 

• Table 1. Why MAE values are lower than bias (MBE) values? In bias (MBE), positive and negative errors cancel out and the resulting MBE value is lower than the one obtained with MAE (summation of absolute errors). In fact, the MBE, MAE, and RMSE should follow these inequalities: MBE ≤ MAE ≤ RMSE.

 

• Table 1. Diameters in DAFB₁₀₄ are larger than in DAGB₁₂₀. I expected the opposite behaviour: an increase of the diameter with time. Please, could you explain what is the reason for this?

 

• Table 2. Some of the F1 values are higher than the corresponding Precision and Recall values. For example, this is the case of DAFB₄₂/W_G-1 (first row) where F1=84.2% while Precision=80.0% and Recall=83.8%. F1 is the harmonic mean of precision and recall, therefore it cannot be higher than both input values. I think that other values ​​in the table should be checked. For example, in the case of DAFB₄₂/W_3-2 (third row) where N=34, TP=30 and Accuracy=74.5%. From these data and by applying the definition of accuracy (Eq. 4), a negative FP value is obtained.

 

• The manually counted fruit number changes so much from one growth stage to another. That is, the total number of fruits is equal to 80, 42, 52 and 78 in DAFB₄₂, DAFB₇₂, DAFB₁₀₄ and DAFB₁₂₀, respectively. The decrease in the number of apples between DAFB₄₂ and DAFB₇₂ could be due to falling apples or thinning operations. But it is more difficult to explain the increase from DAFB₄₂. Please, could you explain the reason?

 

• Line 406. Conclusions: “in the growth stage of DAFB₄₂, enhanced uncertainty (MBE=-38.9 mm)…”. I have not found this MBE value in Table 1 or anywhere else in the results section.

 

• Conclusions should clearly state what it the main contribution of this work. Previous studies have used LiDAR reflectivity and geometric parameters in fruit detection. So, in my opinion, the main novelty is the use of LiDAR sensors to estimate the fruit size. The realization of the study in different growth stages is also a contribution of this work.

 

• As a final recommendation, it would be very interesting to publish your dataset in a public repository. There is a shortage of agronomic datasets and this type of information would be welcomed by researchers working in this field.”

• Minor Comments

  • Line 60. Not all RGB-D cameras are based on ToF principle. Other principles, like active stereo vision are also used.

   

  • Line 250: Please, add respectively at the end of this sentence “above 70’% and below 55%, respectively”.

   

  • Line 312. I guess that a connector is missing in this sentence: “The MAE decreased DAFB₁₂₀…”

   

  • Line 338. A gap is missing before “C”: “Furthermore, the values of C and L...”

   

  • Lines 345-346. “The highest F1 score was acquired...”. To avoid confusion, I suggest to start this sentence indicating that you’re analyzing DAFB₂₄.

   

 

Author Response

Please see the attachement.

Author Response File: Author Response.docx

Reviewer 3 Report

This manuscript reports a LiDAR laser scanner application for apple shape detection. Authors employed geometric and radiometric features in 3D point cloud during different apple fruit development stages, different apple tree height from both before and after defoliation. In general, the topic is very interesting and current, the manuscript written in a professional way with clear language and adequate description for all manuscript sections. I just noticed these 2 points:

L37: Mostly you mean LAI.

Figures 1 & 2: Improve the quality of these flowcharts as some parameters are not clear.

Author Response

Reviewer #3 comments and suggestions

L37: Mostly you mean LAI.

The full name of the term was written. We did not use the LAI, since it has a different meaning.

Added or modified text in the revised manuscript:

Line 39: “the variation in the leaf area”

 

Figures 1 & 2: Improve the quality of these flowcharts as some parameters are not clear.

The quality of the figures was revised and the list of abbreviations updated.

 

Reviewer 4 Report

Summary：The manuscript is about classification of 3D points of apples and measuring the diameter of apple. Some parts of the manuscript read quite well, other parts need improvements to enhance. I have found some issues regarding the practical application of the method proposed by this paper.

i)abbreviations in the article need to be checked.For example, N has different meanings in the manuscript.

ii) There is too little data(Only two trees) to evaluate the apple detection methodology. The DAFB₁₂₀ in section 3.1 and 3.2 is the data of the same two trees? Whether the results obtained from the data analysis of these two trees are applicable to other trees? More data is needed to prove this.

iii) What is the significance of calculating the distance difference between the M_L and M_D? M_L and M_D are both calculated by method proposed in Section 2, and there is no reference value to evaluate this method. Add the actual Center of fruit points to evaluate this method.

 

1、line 23: the full name of"R_adj^2"and "DAFB₁₂₀" should be given, as fist time appeared in the abstract.

 

2、line 37: What does "LA" mean? Is it L_A? It’s full name should be given, as it fist time appeared in the article.

 

3、line 78-80: Change “aimed” to “aiming”.

 

4、line 87：“a geometric factor”should provide specific geometric factor

 

5、line 111-114 and line 124-125: Readers will be clearer if you give the structure diagram

 

6、line 141:Only two trees is too little data to prove the validity of the method

 

7、line 142：What does "N=8" mean？

 

8、line 151-152：“The total number of the P_i set (N)”，The meaning of the N in this sentence is different from the N in the abbreviations，please use different abbreviations to express different meanings

 

9、line 168：What does "N=2" mean？

 

10、line 194: the full name of "R ̃" and "C ̃"should be given, as it fist time appeared in the article.

 

11、line 200：Delete ” and” from ”….of and…”

 

12、line 230: What does "N=6" mean？

 

13、line 232：Is N equal to 6 in formula？

 

14、line 238：“N denotes the total number of measurements”, The meaning of the N in this sentence is different from the N in the abbreviations.

 

15、line 268:What does "R" mean? Is it R_tof?

 

16、line 247-253 and line257-263: It is much clearer to describe the data in a table

 

17、line 305: The full name of "R_adj^2" should be given, as it fist time appeared in the article.

 

18、line 353: The DATA₁₂₀ data used in Table 2 is the data used in the analysis in Section 3.1? Analysis and evaluation using the same data is not rigorous. The evaluation should be done using data that is different from the data used in the analysis.

 

19、line 401: "R_ToF" shou be " R_th"

 

20、line 406: "MBE= -38.9 mm "  is not correct, it is different from the data in the Section 3.

 

21、In general, please double-check abbreviations to ensure uniqueness of abbreviations.

 

22、The efficiency of the algorithm is not mentioned in the article，What is the average processing time of this algorithm and how long it takes to compute a tree？

Comments for author File: Comments.docx

Author Response

Please see attachement.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Apple shape detection based on geometric and 2 radiometric features using a LiDAR laser scanner

Minor Revisions (Rating out of 100 (60=best)) 47(=78%)

(1-10(best))
Novelty: 7
Writing&Language: 9 (+2)
Scientific Rigor: 9
Introduction&Literature: 9(+3)
Method&Results: 9(+1)
Discussion&Conclusion: 4(-2)

In short: Your paper proposes a method to detect apples on trees, using LIDAR.

Expertise of Reviewer: I am not currently performing research in the field of remote sensing; however, I lectured it until recently. I am currently developer and researcher in photogrammetry and LIDAR.

Overall Statement:
Thank you very much for considering my comments. The last PlagScan did not return any plagiarism. Furthermore, I am very happy with how you presented your changes for the second review round.
Unfortunately, you did not consider my (or the other reviewers) comment regarding your conclusion and even extended it largely to a repetition of the results instead of concentrating on the main contribution of your research and whether the aims and objects were achived, which you so clearly present at the end of your introduction section. "The aim of this study is to propose a methodology for segmenting, localizing, and analyzing fruits on the tree, based on the assumption that apples show enhanced RToF at 905 nm compared to foliage and woody parts. The main objectives are: (1) the classification of 3D points of apples based on reflectivity and geometric features derived from LiDAR laser scanner; (2) the development of a fruit segmentation algorithm that allows measuring the fruit size in defoliated trees; and (3) the evaluation of the proposed technique on foliated apple trees."
So, did you achieve to create a method for segmenting, localizing and analysing fruits on the tree, were able to classify them and so on..? This can be done is very few precise sentences and should not be so long. The conclusion is not a repetition of the paper.

Detail Feedback:
Line 233 Figure 2 description: "point out to the", I assume you mean "point out the"

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors have corrected the mistakes and have substantially reorganized the structure of the paper. Overall the article has been significantly improved. I have only detected two formal issues that should be corrected. For the rest, congratulations to the authors for their work.

• There is an error in the numbering of the Discussion and Conclusions sections. Subsections “Segmentation” and “Evaluation” should also be numbered.
• Resolution of Figure 5 should be improved.

Author Response

Please see attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

Summary：Generally speaking, the authors adequately address all of my big questions within their response to reviewers. The current version of the manuscript is significantly improved. I’ve listed a number of comments below but in general I recommend it for publication in Remote Sensing.

1、line 66-68: This reference paper present “similar” algorithms in apple shape detection and resulting in 0.89 F1-score and 94.8 % average precision. The result seems to be better than the method proposed in the paper(The algorithm resulted in maximum values of 88.2 % precision, 91.0 % recall, and 89.5 F1 score at DAFB₁₂₀). What are the advantages of this article compared to this reference article.

2、lin255-261：In my opinion, these data are hard to come by just looking at Figure 3. How did you get the data for the different parts, which need to introduce the process of obtaining these data in the article.

3、line290-299: These threshold ( RToF > 76.1 %, L < 15.5 %, and C > 73.2 %) were analyzed from DAFB₁₂₀ data. From the results in Table 1, These threshold may not be applicable to DAFB₄₂ DAFB₇₂ DAFB₁₀₄. Would it be better to get different thresholds by analyzing the data at different DAFB?

4、These threshold ( RToF > 76.1 %, L < 15.5 %, and C > 73.2 %)  are obtained from two trees and whether they apply to other trees? When the method in this article is applied to other trees in the orchard, do thresholds need to be recalibrated?

Comments for author File: Comments.docx

Author Response

Reviewer #4 comments and suggestions

• line 66-68: This reference paper present “similar” algorithms in apple shape detection and resulting in 0.89 F1-score and 94.8 % average precision. The result seems to be better than the method proposed in the paper (The algorithm resulted in maximum values of 88.2 % precision, 91.0 % recall, and 89.5 F1 score at DAFB₁₂₀). What are the advantages of this article compared to this reference article.

The advantage of this study is its calibration method. We scanned 2 trees pre and after defoliation at each growth stage. The defoliated tree samples increased the resolution of the thresholds. The detected apples were used as ground truth labels, while their size (D_D) was also evaluated over the manual measurement (D_Manual). In Gene-Mola et al., 2019, it was assumed that the area of high intensity value, in foliated trees, can possibly belong to apple. These areas were manually cropped in the foliated 3D point cloud and used as labels for the further analysis.

 

• Line 255-261：In my opinion, these data are hard to come by just looking at Figure 3. How did you get the data for the different parts, which need to introduce the process of obtaining these data in the article.

The first paragraph describes the range of R_ToF, L and C for each class (wood, leaves, apples). In Line 266-274, we mention the significance of the defoliated tree for readjusting the range from apples and woody parts. The range of leaf class for the same features was defined in the foliated trees. Whereas latter we used the probability density function to acquire the corresponding thresholds.

 

• line290-299: These threshold ( RToF > 76.1 %, L < 15.5 %, and C > 73.2 %) were analyzed from DAFB₁₂₀ From the results in Table 1, These threshold may not be applicable to DAFB₄₂DAFB₇₂ DAFB₁₀₄. Would it be better to get different thresholds by analyzing the data at different DAFB?

The results showed (Table 1, 2) that the use of threshold values from DAFB₁₂₀ can segment and detect the size and the number of apples per region at different growth stages. We acknowledge the fact that the results could be improved by implementing the corresponded threshold values at each stage. However, we decide to not over manipulate our algorithm.

 

• 4、These threshold ( RToF > 76.1 %, L < 15.5 %, and C > 73.2 %)  are obtained from two trees and whether they apply to other trees? When the method in this article is applied to other trees in the orchard, do thresholds need to be recalibrated

This is something that needs to be further investigated. Currently we are working on the further development of the method using a bigger set of trees to calibrate the thresholds and training set. For the upcoming years we would check the robustness of the thresholds obtained with calibrated laser scanner.