A Deep Learning-Based Method for Extracting Standing Wood Feature Parameters from Terrestrial Laser Scanning Point Clouds of Artificially Planted Forest
Round 1
Reviewer 1 Report
The presented research proposes an approach for tree feature parameter extraction from point clouds. The topic is timely and significant.
The study is correctly designed and scientifically sound in general. However, some parts should be presented more clearly. Below, you can find some suggestions that in my opinion will help improve the manuscript.
Line 19 – do not use the term terrestrial laser scanning lidar, just terrestrial laser scanning
Line 50 – backpack scanning systems, and mobile mapping systems (on vehicles) are also terrestrial laser scanning systems, just not static but kinematic… please revise the sentence
Line 52 – in this case we are not talking about a remote sensing application
Line 102 – please check the sentence, and rephrase
At the end of the first section, it would be appropriate to highlight the advantage of your approach. Maybe to compare it with the disadvantage of other methods.
Line 128 – please explain why the color or the intensity is omitted, eventually what does it mean that you segment the points that miss the color and the intensity because the result of TLS is XYZ and intensity (and true color if photos are taken)
Line 150 – the stated accuracy is defined as one sigma at 100 m so for the range of 2500 m it is much higher approx. 125 mm, this fact is important because of the interpretation of the results and the accuracy of your approach.
Line 218 – please explain the human shadow… is it some points that are not lying on the surface of the trees such as scanned people?
Line 223 – please explain this step in more detail
Fig 5. – check the legend
Equations in the 2nd section – please check the explanation of all the parameters… e.g line 310 the regularization factor or the edge weight – explain their function, etc.
Line 406 – please consider adding at least a basic description of the method
Line 425 – please explain how the value of the overall accuracy is estimated and what it represents
Fig. 9 – add the explanation of each part also into the legend (title) of the figure
Fig. 11. – change the size of the text and replace the “terrain” with “ground” as you are using this term in the rest of the paper
Table 2 – describe in the previous text, just briefly, how the values were calculated
Line 491 – SLAM - there is no info about the use of SLAM in the previous text. What kind of SLAM was used and for what exactly? You used a TLS on a tripod, so basically, you need to register the data into one coordinate system. Please also describe the approach used for registering.
Line 492 – 498 – why is the difference higher in the case of “diameter at breast height”? I am not sure about your explanation that it is because of the inaccuracy of the SLAM (registration I guess), because the manual measurement and the measurement performed by QSM algorithm were executed on the same dataset, with the same “errors”.
Line 553-556 – it would be relevant to compare the methods on the same data because as you write, the type of vegetation affects the accuracy
Author Response
Dear Reviewer
Thank you so much for your suggestion, here is a point-by-point response that has been uploaded and we hope you will be satisfied with our response.
Best Regards,
Xingyu Shen, Qingqing Huang*, Xin Wang, Jiang Li and Benye Xi
Author Response File: 
Author Response.docx
Reviewer 2 Report
An effective method based on energy segmentation and PointCNN is proposed in this work. Experimental results show the good performance of the proposed method. However, some issues should be addressed.
Major issues:
1) The structure of the paper is confused. The experimental data and the proposed method are placed in the Section 2, which is not reasonable. It is suggested that Section 2 introduces the experimental data. The Section 3 introduces the proposed method.
2) In addition, operation time is also one of the important indicators. It is suggested that the authors add the evaluation index of operation time
Minor issues:
1) The methods proposed by the authors is mainly based on image processing. In the introduction part, some image processing techniques should be further introduced, e.g.,
[1] Super-Resolution Mapping Based on Spatial-Spectral Correlation for Spectral Imagery [J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 59(3): 2256-2268.
[2] A Simple Method of Mapping Landslides Runout Zones Considering Kinematic Uncertainties, Remote Sensing, 2022, 14(3): 668.
2) In addition, there are some grammatical errors in the article, which need further careful proofreading.
Author Response
Dear Reviewer
Thank you so much for your suggestion, here is a point-by-point response that has been uploaded and we hope you will be satisfied with our response.
Best Regards,
Xingyu Shen, Qingqing Huang*, Xin Wang, Jiang Li and Benye Xi
Author Response File: Author Response.docx
Reviewer 3 Report
Deep learning is currently one of the most widely researched areas of machine learning. paper presents an efficient process for extracting standing wood feature parameters from the terrestrial laser scannig.
My questions:
1. How many station have been used to provide this area? what is the accuraccy of registration?
2. how many training and validation datasets for the two datasets due to their different characteris were generated?
3. Table 1. : Number of scanned points; NPPT: Average number of scanned points per tree - I think that NPPT sholud be named in percentage, these numbers say nothing about statistics.
However, as Authors claim: there are still some limitations to this work. For example, the manual labeling of point clouds remains highly subjective.
Author Response
Dear Reviewer
Thank you so much for your suggestion, here is a point-by-point response that has been uploaded and we hope you will be satisfied with our response.
Best Regards,
Xingyu Shen, Qingqing Huang*, Xin Wang, Jiang Li and Benye Xi
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
Dear authors,
I think the revised version of the manuscript can be accepted for publication. I wish you all the best.
