A Method Based on Improved iForest for Trunk Extraction and Denoising of Individual Street Trees

Round 1

Reviewer 1 Report (Previous Reviewer 2)

Thanks to the authors for addressing my comments. and Congratulations on your nice work. 

We always recommend to our researchers share the necessary data and codes to increase methods' clarity and reproducibility. Therefore, the methods paper should have contained a sample set of datasets and necessary codes so that others can easily reproduce the results and can solve their problems. Which also increases the clarity of the method. 

I hope the authors must be shared their sample dataset with the methods paper.  

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report (New Reviewer)

The paper is well written and shows good results in extracting and denoising the trunk. However, it still works with traditional approach of identifying the geometric features with statistical reasoning and improving the algorithm. In current scenarios, this should be handled with deep learning algoritm to classify and differentiate between trunks and crown.

 

The methodology section needs to improved with flowchart diagram for more clarity.

There needs to be more information with regards to dataset used. And how much variation exists in the trunk and crown of these dataset needs to be reported.

 

 

The result section needs to clearly explain why there is so much difference between the precision in the first two dataset. And how those scenarios can be avoided. 

The comparison must be reported with other recent established methods not only with different stages of the methodology.

 

The conclusion section can be improved.

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

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 manuscript improved the iForest method on the basis of the previous work in order to expect a fine extraction of tree trunks for the reference of 3D modeling of street trees and their applications. The article has done some work, but there are some problems in the paper that need to be reviewed.

 1.       The introduction section needs to be strengthened and improved. firstly, the literature is not covered enough, secondly, the focus of this paper is to improve the denoising method but the introduction section lists the trunk extraction and 3-D reconstruction methods, and the denoising is less involved.

2.       What do the horizontal and vertical coordinates of Fig. 2 c and Fig. 3 represent? The units should been given. 2.2.1 should state the width of the slice and the reference. In 2.2.2, it is suggested that the width of the slice is 0.05m, why? According to the article, it should be the width of the slice after removing the canopy, please distinguish the two and make the statement clearer.

3.       From Figure 3, we can see that the minimum RMSE is 0.004 but the RMSE of the optimal slice in Figure 2 is 0.009, so it is recommended to keep the same. At the same time, what is described in Figure 2 should be the slice of the trunk and the fitting result, please express the figure name clearly, or add the fitting schematic of the tree crown slice in the figure.

4.       The paragraph above Figure 5 says that the BIRCH method can separate the target point cloud from the outlier points, thus achieving the purpose of separating the discrete points. However, Figure 5 shows the result of the BIRCH method, which only reduces the number of point clouds while maintaining the trunk shape, and does not separate the outlier points.

5.       The manuscript only provides a comparative methodological analysis about the results of Figure 5, a and c. what about the cases of Fig5 b and c? Please add. And please explain the problem of missing contour point cloud in case figure5a.

6.       What does “total” stand for in Table 1? For the overall accuracy of cases 1-4 or the average accuracy of the 4 cases, it is not stated in the text and the table.

7.       Many variables in the charts are not clear, and it is suggested that the meaning of each chart can be read directly from the chart name. Also the chart should be placed after the paragraph in which it first appears in the text, in addition to editorial layout interference.

8.       The description of Figure 10 is three scenes and four datasets, but dataset 4 belongs to different scenes of the campus, so the description of three scenes does not match well. And the last sentence in this paragraph is "… the three datasets have 14,29,19 and 14 trees respectively" which is easy to be misunderstood, so it is suggested to explain clearly.

9.       The abbreviated expressions should appear first in the manuscript, followed by their results, such as Eqs. 9 and Table 1, and Table 1 should appear after Eqs. 9 and 10. The same should be done for the rest of the text. It is suggested that the author should adjust the order of text description. Section 3.3 should be placed in the method section, etc, please think about it.

10.   Which case in Figures 11-13 corresponds to Figure 14c? Where is the result of the correction for the error extraction in Figure 12? Is the order of Figures 14 and 15 correct? How is the error separation trunk and its noise improved in Figure 15?

11.   10. Not all cases of error extraction and noise removal have been analyzed in the discussion, e.g., only 11c has been discussed, but the cases of d and e still need to be explained.

12.   "…the proposed method is also applicable to tree trunk extraction in other scenes." "In addition, the proposed denoising method is also applicable to the denoising of other pole-like objects." The text does not mention other scenes except for the street and the campus, nor does it experiment with other pole-like objects. It cannot be stated that this method is applicable to other scenes or objects.

13.   "…the denoising accuracy of the proposed method can be improved by approximately 30% for noise points tightly close to tree trunks". This is the result of the method comparison based on a single tree and is not stated in the text. In addition, in the 3 scenes of the example also did not reflect the accuracy can be improved by 30%, it is recommended to increase the method comparison experiment.

14.   In this paper, IForest is improved, and it is suggested to add the comparison with IForest method.

15.   The format of the article needs to be improved, and some English descriptions need further polish.

Reviewer 2 Report

The author wrote an interesting manuscript but needs some improvement. Specifically, the sample size is low (72 trees). Methodological papers need to be more sampled. Street trees are mostly found isolated or straight, which gives them available space to separate in point clouds. I would love to see more sampling from structurally diverse trees based on dominance level. However, maybe the following suggestions would help authors to improve their manuscript.

Abstract:

Why do street trees have excessive noise? Due to wind?

What type of noise patterns do the authors solve with their "iForest algorithm" that is not mentioned in the abstract?

Slicing seems to me always problematic due to the chance of overlapping portions, how do authors avoid such difficulties?

Methods:

I would recommend authors separate or write study sites, with methodology (in separate sections), not with results.

I think the sampling or tree selection procedure should be improved as most of the trees are selected from denser canopies, but how does this algorithm work for lean trees or overlapped stems?

Discussion:

The major concern for me is why and how noise happens, and how this algorithm solves that problem is missing. I recommend authors add a discussion and an introduction.