A Multiscale Multi-Feature Deep Learning Model for Airborne Point-Cloud Semantic Segmentation
Round 1
Reviewer 1 Report
I congratulate the authors on the work done. For publication, a few corrections need to be made to the figures and the markings used.
Numerous abbreviations and designations appear in Figure 2. A legend should be attached to the drawing with the used designations.
Figure 3 in part b shows the different sizes of the drawings and it seems that a different scale is adopted. Does part a have the same parameters adopted as part b?
The caption in Figure 8 should include parts a, b and c.
Punctuation should be improved throughout the paper. It is mainly about editorial issues: unnecessary periods, lack of spaces after periods, etc.
Author Response
Dear Dr. Reviewer:
We sincerely express our appreciation for your time and effort in reviewing and commenting on our manuscript entitled “MSMF-PointNet:Semantic Segmentation of Airborne LiDAR Point Cloud based on PointNet Fusion with Multiple Features” with the manuscript number (applsci-1998876). We found the comments are very valuable and helpful when revising and improving the manuscript. We have responded to the comments and made the necessary corrections, which have been marked in RED in the text. We hope that the revision meets your approval.
The comments along with the corresponding changes are as follows:
Point 1:Numerous abbreviations and designations appear in Figure 2. A legend should be attached to the drawing with the used designations.
Response:Thank you for your valuable comments, we have added the appropriate notes below Figure 2. The content added is shown below.
(R=0.8m and R=102m: the radius of the spherical neighborhood is 0.8 and 1.2 respectively; R, O, P, L, V: Roughness, Omnivariance, Planarity, Linearity, Verciticality.)
Point 2:Figure 3 in part b shows the different sizes of the drawings and it seems that a different scale is adopted. Does part a have the same parameters adopted as part b?
Response:Part a have the same parameters adopted as part b, The two images in b of Figure 3 are in the same scale. We replaced the image with a higher resolution in the manuscript, and we apologize for the trouble we made in creating the image.
Point 3:The caption in Figure 8 should include parts a, b and c.
Response:Based on your suggestion, we have added the appropriate content to the title of Figure 8. The content added is shown below.
((a) represents a zoomed-in image of the misclassification details of the model proposed in this paper on Shrub, (b) and (c) represent the zoomed-in details of the misclassification of the model proposed in this paper on Tree.)
Point 4:Punctuation should be improved throughout the paper. It is mainly about editorial issues: unnecessary periods, lack of spaces after periods, etc.
Response:Thank you for pointing this out, and we have checked the punctuation throughout the text to correct it.
Author Response File: 
Author Response.docx
Reviewer 2 Report
The study seems to be a good research work, however, the following are my comments on it:
1. Title of the study is good but there's a sense of unclarity in the sentence structure. For example, a clear sense of the title can go like this:
A Multi-Scale Multi-Featured Point Cloud Deep Learning Model for Semantic Segmentation of UAVs
A Multi-Scale Multi-Featured Deep Learning Model for Point Cloud Semantic Segmentation of UAVs
However, these are just the ideas, and authors are advised to come up with a more clear and more crispy title for the study.
2. In addition to comment 1, the sentence structure needs to be looked at in the abstract as well. Long sentences tend to get vague and lose the readiness of the authors along the way. For example, in the abstract, a sentence starts from line 17 and goes to line 21. Look into this matter.
3. In line 35, a study cited with authors doesn't match the cited paper in the citations. In line 37, the authors mentioned a new study but didn't cite it. This can badly affect the importance and effectiveness of the study. Authors are advised to look into this issue for a complete paper. In a similar sense, the section needs to be revised for better sentence structure and to get it clear with the notion it wants to address. A similar issue is persistent on line number 44.
4. The introduction section provides a compact view of the problem domain with clear contributions to the study. However, the current section provides no such information. This can affect the readiness of the study for its reader to grab the crisp of the study.
5. The proposed set of features are have been used widely in many existing studies. Authors have not provided a justification of why they have considered such feature engineering mechanisms when there is a wide spectrum of other mechanisms that exist.
6. Similarly, it is better to provide the architecture of model layers as well as a training and testing process to reproduce the study by new researchers. The image conversion to 2D and reconstruction of 2D to 3D processes are also missing. Authors must include the algorithmic process of feature engineering and training of proposed models.
7. The reference studies are a bit older. The authors are advised to review the new literature to address the recent state-of-the-art studies.
Author Response
Dear Dr. Reviewer:
We sincerely express our appreciation for your time and effort in reviewing and commenting on our manuscript entitled “MSMF-PointNet:Semantic Segmentation of Airborne LiDAR Point Cloud based on PointNet Fusion with Multiple Features” with the manuscript number (applsci-1998876). We found the comments are very valuable and helpful when revising and improving the manuscript. We have responded to the comments and made the necessary corrections, which have been marked in RED in the text. We hope that the revision meets your approval.
The comments along with the corresponding changes are as follows:
Point 1: Title of the study is good but there's a sense of unclarity in the sentence structure. For example, a clear sense of the title can go like this:
A Multi-Scale Multi-Featured Point Cloud Deep Learning Model for Semantic Segmentation of UAVs
A Multi-Scale Multi-Featured Deep Learning Model for Point Cloud Semantic Segmentation of UAVs
However, these are just the ideas, and authors are advised to come up with a more clear and more crispy title for the study.
Response: Thank you for your valuable suggestion, we decided to modify the title according to your suggestion as follows：A Multi-Scale Multi-Featured Deep Learning Model for Airborne Point Cloud Semantic Segmentation.
Point 2: In addition to comment 1, the sentence structure needs to be looked at in the abstract as well. Long sentences tend to get vague and lose the readiness of the authors along the way. For example, in the abstract, a sentence starts from line 17 and goes to line 21. Look into this matter.
Response: To make the meaning of the sentence clearer, we rewrote the sentence in the summary, which is located in lines 18 to 21 of the manuscript. The revised contents are as follows.
In this paper, we use the spherical neighborhood method to obtain the local neighborhood features of the point cloud, and then we adjust the radius of the spherical neighborhood to obtain the multi-scale point cloud features. The obtained multi-scale neighborhood feature point set is used as the input of the network.
Point 3: In line 35, a study cited with authors doesn't match the cited paper in the citations. In line 37, the authors mentioned a new study but didn't cite it. This can badly affect the importance and effectiveness of the study. Authors are advised to look into this issue for a complete paper. In a similar sense, the section needs to be revised for better sentence structure and to get it clear with the notion it wants to address. A similar issue is persistent on line number 44.
Response: Thank you for pointing out the problem in this article; the citation error in line 35 has been corrected in the manuscript. We have divided the new research mentioned in line 37 into three parts in the following section (three paragraphs from line 41 to 81). In addition, we have moved the insertion of the quotation from the end of the sentence to the middle of the sentence, which will make the text flow better. The problem in line 44 has been corrected.
Point 4: The introduction section provides a compact view of the problem domain with clear contributions to the study. However, the current section provides no such information. This can affect the readiness of the study for its reader to grab the crisp of the study.
Response: Thank you for your valuable suggestions. We have added the corresponding content to the current section of the manuscript(From line 82 to line 93). The specific additions are as follows.
In the current research on semantic segmentation of point clouds, If it is difficult to achieve fine classification of point clouds only by relying on sparse three-dimensional coordinates of airborne point clouds,Li et al.[21] designed a refined feature extractor using self attention mechanism to improve the accuracy of point cloud classification; Yang et al.[22] proposed a graph attention feature fusion network (GAFFNet) that can achieve a satisfactory classification performance by capturing wider contextual information of the ALS point cloud. Luo et al.[23] confirmed the potential of multispectral LiDAR in the classification of complex urban land cover through three comparison methods. multispectral LiDAR has the availability of spectral information; Li et al.[24] have constructed feature pyramids to integrate features at different scales and thus classify point clouds with good results. However, these methods suffer from the problem of excessive memory consumption or computational consumption.
Point 5: The proposed set of features are have been used widely in many existing studies. Authors have not provided a justification of why they have considered such feature engineering mechanisms when there is a wide spectrum of other mechanisms that exist.
Response: Thank you for your suggestion. We have rearranged and added to the manuscript the rationale for the selection of these features (from lines 169 to 191). The additions are as follows.
λ1, λ2, and λ3 are eigenvalues of the point cloud, where λ1≥λ2≥λ3. An analysis of the eigenvalues and eigenvectors can often provide important information for extraction decisions. According to the points in the neighborhood, the covariance matrix of the center point was calculated,and then the eigenvalues of the point were obtained[27]. Based on these eigenvalues, 4 kinds of features can be calculated,including sum of Omnivariance(Oλ), Planarity(Pλ), Linearity(Lλ), Verciticality (Vλ).
Oλ has a strong ability to describe the fluctuation degree of point cloud surface. Generally speaking, complex surfaces have a higher Oλ values. The Omnivariance of trees and grass is higher than that of man-made surfaces and buildings.
Pλ is a measurement of planar characteristics of the point cloud, It can effectively represent the level of the fitted surface in the neighborhood at this point. The flatter the surface is, the higher the flatness will be. For example, the flatness of the artificial road surface is significantly higher than that of the tree surface.
Lλ denotes the degree of linearity of the point cloud. The power lines and edges of buildings have obvious linear structures,and the linearity of these points is characterized by high values.
As for Vλ,within 90 degrees, the larger the angle between the surface and the ground is, the higher the Vλ of the points on the surface of the surface of the surface will be. The Vλ of tree trunks, walls, street lamps and fences will be higher, while the road surface and grass will be obviously lower.
Compared with a single feature, the combination of all the above features can provide more effective information for subsequent classification and achieve better results.
Point 6: Similarly, it is better to provide the architecture of model layers as well as a training and testing process to reproduce the study by new researchers. The image conversion to 2D and reconstruction of 2D to 3D processes are also missing. Authors must include the algorithmic process of feature engineering and training of proposed models.
Response: Thanks to your suggestion, we have uploaded our code on GitHub. We hope it will be beneficial for new researchers to reproduce it. https://github.com/fmq1984/MSMF Our images and point clouds are obtained directly by scanning. On the process we do the convolution with the point cloud in xyz representation.
Point 7: The reference studies are a bit older. The authors are advised to review the new literature to address the recent state-of-the-art studies.
Response: Thanks to your suggestion, we have added four articles of this two-year literature between lines 84 and 91 in the manuscript. The specific additions are listed below.
[21] Y. Li and J. Cai, "Point cloud classification network based on self-attention mechanism," Computers and Electrical Engineering, vol. 104, 2022, doi: 10.1016/j.compeleceng.2022.108451.
[22] J. Yang, X. Zhang, and Y. Huang, "Graph Attention Feature Fusion Network for ALS Point Cloud Classification," Sensors (Basel), vol. 21, no. 18, pp. 6193-6193, Sep 15 2021, doi: 10.3390/s21186193.
[23] B. Luo et al., "Target Classification of Similar Spatial Characteristics in Complex Urban Areas by Using Multispectral LiDAR," Remote Sensing, vol. 14, no. 1, pp. 238-238, 2022, doi: 10.3390/rs14010238.
[24] D. Li et al., "AGFP-Net: Attentive geometric feature pyramid network for land cover classification using airborne multispectral LiDAR data," International Journal of Applied Earth Observation and Geoinformation, vol. 108, 2022, doi: 10.1016/j.jag.2022.102723.
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
I appreciate the efforts of authors but there are still some problems in the manuscript.
1. Author must improve the english and typos. I gone through some of the sentences that must be rephrased. Some sentences are long and unclear to understand.
2. Most of the references are still too old. As you are publishing your work in the end of 2022. You must cite the work from 2020, 2021, and 2022. but your many citations are old.
3. Add organization of the paper in the end of the introduction section.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
