Point Cloud Data Processing Optimization in Spectral and Spatial Dimensions Based on Multispectral Lidar for Urban Single-Wood Extraction
Round 1
Reviewer 1 Report
Thanks for your patience while I reviewed the article. Unfortunately the article is not innovative enough to be accepted for publication in a Q1-ranked journal. Also, The language and structure of the article are poor. Have a native English speaker read the paper . I hope you find a good home for it in another journal.
Author Response
Please see the attachment
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The method presented in this manuscript has some novelty, and the experiments show that it could provide better results. However, the presentation should be improved before publication. Please see the following details.
Line 64: ability of estimating …
Line 119: “…was displayed by …”
Line 121: Table 2 should be Table 1?
Line 147: … and the blue point cloud …
Figure 4: what different colors refer to should be provided either in the caption or legend.
Line 223: “Table 1” should be “Table 2”.
The formulas are not numbered.
Some of the statements (e.g. lines 281-303) in section 3 are method descriptions rather than results, which could be moved to section 2.
Line 277: the specific total classification accuracies before and after processing should be reported.
Line 318: I suggest revising this statement as: According to the experimental results of joining all the spectral indices, …
Line 333: were the trees divided manually or automatically by a segmentation algorithm?
Table 7: I guess 308 should be named as field-surveyed trees and the other numbers should be named as Lidar-detected trees. Meanwhile, OA in Table 7 is actually the detection rate (or recall) of tree segmentations, which is related to errors of under-segmentations. In tree segmentation, both under-segmentation and over-segmentation errors should be evaluated. I suggest using recall, precision, and F-score to evaluate the single tree extraction results.
Discussion section: similar studies using multiple-spectral Lidar could be compared. Another question is whether multiple-spectral Lidar data can provide significant improvements over single band Lidar data in urban point cloud classification and urban single wood extraction. This could also be discussed, and some experimental results could be added.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 3 Report
The article deals with the optimization of point cloud classification obtained from LIDAR technology. The proposed algorithm is based on a combination of single- and multispectral data using different types of vegetation indices. The method was verified on multispectral airborne Lidar Optech Titan data of 1550, 1064, and 532 16 nm channels. The modeled area was an urban area around the University of Houston in Houston, Texas, US. Although the article is well structured and the method sufficiently explained, I would like to make a few suggestions that could improve this publication.
In my opinion, the description of the current state of the issue is insufficient. The article mainly evaluates LIDAR data technologies, but the list of related works should be supplemented with research that is directly related to the classification and evaluation of LIDAR data. It is clear from the text that it is a supervised classification using the k-means method. Is this method universal enough? If I change the modeled area, do I have to generate new training samples or are these samples applicable to other areas as well? It is not clear from the text how and in what software the samples were tested. Furthermore, information is missing in which software the proposed optimization algorithm was implemented. Is it commercial or custom software? In the scope, the proposed method should be compared with the existing solution (related works) and the advantages of the proposed solution should be clearly demonstrated.
After completing this information, I recommend the article for publication.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 4 Report
The authors demonstrated the data processing method of a three-wavelength Lidar for urban single wood extraction. Since the device is already commercialized and the data processing methods adopted in the manuscript are quite sophisticated, I suggest the authors clearly address the differences or uniqueness of their contribution in the introduction part.
For the technical contents, I also have some suggestions for the authors to consider during their revision.
1. The authors have published a chain of papers on multispectral Lidar. Some of them might use multi-spectral Lidar with two or three wavelengths and some might use supercontinuum laser sources. Can you discuss the differences and limitations of these systems in the introduction part or discussion part?
2. Another possible solution to urban single wood extraction is the combination of passive spectrometer and one-channel Lidar. In this scenario, the classification can be achieved by the spectral data and be further verified by the 3D modal (based on the ground and canopy features) reconstructed by a Lidar. Please compare and discuss this in the revised manuscript.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 5 Report
Dear Authors
The topic of the paper (3D segmentation and its classification) is new and the structure is completed. The author and his team write this paper according to journal scope and modern trends. The novelties of the paper are clear. One is to add a preprocessing on the cloud segmentation (this new approach is your major advance) and the other is to extract single wood with down sampling. I hope the authors can focus on the following two issues.
1. After applying the preprocessing, the accuracy increases by 3.7%, I do not think that the gap between new method and conventional method is very big. And the authors did not mention the exact accuracies with and without the method, respectively. You may persuade others that the differences are not random. If it is possible, you may split your data into 4 pieces evenly. You can conduct your new methods on all of the four. If each of them can achieve an increase with the new method, it can be plausible for the author. If you have other approaches to explain your effectiveness of your method, you can use your own way to verify it.
2. You classify the six groups with spectral indices. I found out that some of them were very similar with others. In Table 5, the accuracy of pure spectral information is 82.22% and the accuracy of all spectral and all indices is 85.12%. You can add some explanation why there is not overfitting in your result. Is that possible that you can choose some of indices but not all of them?
The following are the specific issues in your paper. There are still some grammar problems around the paper. I hope the author can read through the paper for a few times to avoid some errors.
Line 82 “we appropriate” What does it mean?
Line 108 There is a space.
Line 110 “titan” Does the word have an upper case, like Titan
Line 119 “All point cloud display by cloudcompare software”. Should we use the passive voice?
Table 1. You divide the building into two types: housing and Non-residential building. You can explain whether there are some differences between these two types like roughness, texture or geometric characteristics.
Line 154. The whole sentence does not make it sense.
Line 168-176, you try to explain the algorithms. Can you put them after the algorithm? Now you put them in front of the algorithms.
Line 169. You use “was”. But the whole paragraph is the present form. Make it consistent. The problem also occurred on other place and correct them.
Line 195-201. The sentence does not make sentence. You can split them into a few sentence.
Line 220 maximum not maximal
Line 285 Other numbers in your paper does not have comma. But this number has comma. Just make all your numbers consistent in one paper.
Line 402 You can replace studies with other words. It does not make sense.
Table 4 and Table 5. Some of them are lower cases and some of them are upper cases.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
The manuscript has been improved, and my previous major concerns have been addressed. Some minor issues still exist.
Line 193: …, and the points of 532 nm… A space is necessary before the unit nm.
Table 8: if the number in the third column (e.g. 76.2%) is evaluated without considering spatial location, then it should not be called recall. Try to use a different name and give detailed definition.
Line 496: single channel lidar.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
The authors have revised according to my suggestions and I am pleased to give my recommendation of acceptance to this revised manuscript.
Author Response
Thanks a lot for the valuable comments from you. We have revised our paper very carefully under those comments and suggestions which helped us a lot to improve the quality of this paper.
Reviewer 5 Report
I agree to accept it.
Author Response
Thanks a lot for the valuable comments from you. We have revised our paper very carefully under those comments and suggestions which helped us a lot to improve the quality of this paper.
