Research on Low-Altitude Aircraft Point Cloud Generation Method Using Single Photon Counting Lidar

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

In the article titled “Research on Low-Altitude Aircraft Point Cloud Generation Method Using Single Photon Counting Lidar” the authors Zhiganag Su et al. propose an aircraft point-cloud generator based on conditional GANs. They seek to address the issue of lack of quality data acquired via single photon counting lidar for low altitude aircraft identifiers using this approach. 

Although the work is mostly focused on research outside my area of expertise, I found the paper to be well written and would be happy to recommend it for publication with the following minor revisions:

• PCAG is first used in the abstract, but not defined until much later.
• Typo in line 344: “attitude” should be “altitude
• How is the number “1.8” arrived at in the equation (10)? It will be helpful to mention that in the paper.
• Fig. 8: why does background show up as -1? Shouldn’t it be white (>>1, approaching infinity)? (Not necessary to change the figures, maybe just make a comment about how "no return" shows up as black.)
• In the results section we can see from Fig.14 that the proposed method performs better than the alternative. Considering the detailed nature of the article, it might be useful to quantify how much better the approach works with and without Norm.
• Labelling in some of the figures eg. Fig. 13 is very hard to make out. Font size should be adjusted appropriately.
• Also Fig. 13: wouldn’t moving helicopter blades average out to a disc?
• Line 576: it looks like one power of 10 is missing from the expression

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

This paper proposes a novel approach for generating aircraft point clouds in low-altitude airspace under the Single Photon Counting Lidar (SPC-Lidar) system. The method, named PCAG (Point Cloud of Aircraft Generation), utilizes an improved conditional Generative Adversarial Network (cGAN) framework to generate high-fidelity point cloud data. The core of PCAG is an aircraft depth image generator, trained via adversarial learning on simulated point clouds derived from the ModelNet40 dataset. The generated point clouds exhibit high similarity to real-world Lidar-collected data and demonstrate robustness against variations in slant range and aircraft attitudes.

Overall, the paper presents a well-structured study with significant contributions, but addressing the identified limitations would improve its impact and applicability.

MA:

1) Some grammatical inconsistencies, awkward phrasing, and long-winded sentences affect readability. I suggest the author revise the manuscript and improve the fluency and readability.

2) Introduction: It’s weird to make a summary of contributions in the introduction. Please move this part to conclusion or discussion.

3) Lack of Real-World Validation: The generated point clouds are evaluated using simulated data, but no comparison with actual SPC-Lidar measurements is provided. Real-world validation would enhance the credibility of the method. Please address this issue in discussion.

4) Fixed Aircraft Types: The training dataset is derived from ModelNet40, which limits aircraft diversity. Generalization of various aircraft models remains uncertain.

5) Computational Efficiency: The paper does not discuss the computational complexity of PCAG. Since deep learning models, especially cGANs, require substantial computational resources, performance benchmarks should be included.

 

Minor:

• The manuscript includes 7 subtitles from introduction to conclusion, which make the structure complex. Some of the subtitles should be removed or merged to make the structure briefer and easier to read.

Comments on the Quality of English Language

Some grammatical inconsistencies, awkward phrasing, and long-winded sentences affect readability. I suggest the author revise the manuscript and improve the fluency and readability.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

• This paper presents an innovative approach to generating low-altitude aircraft point clouds using Single Photon Counting Lidar (SPC-Lidar) and an improved Conditional Generative Adversarial Network (cGAN). The proposed PCAG method delivers promising results, showing strong accuracy in generating adaptive point cloud data.
• The methodology is well-explained, and the experimental results demonstrate high similarity between the generated and actual point clouds. However, discussing computational efficiency and potential real-world challenges, such as processing time and hardware constraints, would strengthen the study.
• It would be helpful to explore how well the method scales across different aircraft types, altitudes, and environmental conditions. Addressing limitations, such as noise interference and adaptability to different SPC-Lidar systems, would add more practical insights.
• Future Work: Expanding the model to integrate multi-sensor fusion (e.g., SPC-Lidar with radar or optical imaging) could enhance detection capabilities. Exploring real-time processing feasibility and optimizing the model for onboard applications would make it even more impactful.
• Overall, this is a well-executed study that makes a valuable contribution to remote sensing and aerial point cloud generation. With minor refinements, it will be even stronger.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

Comments and Suggestions for Authors

Thank you for inviting me to review the manuscript titled “Research on Low-Altitude Aircraft Point Cloud Generation Method Using Single Photon Counting Lidar”. This paper proposes a conditional GAN-based framework (PCAG) for generating aircraft point clouds under SPC-Lidar systems, offering a novel solution to the scarcity of training data in low-altitude scenarios. While the method demonstrates innovation in dynamic target adaptation, several aspects require clarification and enhancement. Below are my detailed comments:

1. Novelty and Contributions

• Transforming point cloud generation into a conditional depth image synthesis task, which effectively handles six-degree-of-freedom aircraft poses.
• The integration of self-attention layers and Wasserstein loss with gradient penalty improves generation diversity and stability.
• The Norm. module successfully mitigates slant range variations, showing promise for real-world deployment.

Suggestion: Further clarify the technical boundaries between PCAG and image-based methods (e.g., NeRF) in the Discussion.

2. Experiments and Analysis

• Strengths:

Quantitative results (e.g., 12% MMD reduction and 18% JSD reduction) validate the high similarity between generated and reference point clouds.

Robust performance across 3–18 km slant ranges proves the method’s adaptability to dynamic scenarios.

• Weaknesses:

Only 8 aircraft types from ModelNet40 are tested, omitting critical low-altitude targets (e.g., drones).

Computational time per frame is not reported, hindering real-time application assessment.

Suggestions:

1.      Extend the dataset to include drones and evaluate generalization capability.

2. Provide quantitative metrics on generation speed and GPU memory consumption.

3. Method Comparison and Technical Depth

• Comparisons are limited to voxel-based methods and Tree-GAN, ignoring state-of-the-art approaches (e.g., Diffusion models).
• The impact of depth image resolution on point cloud quantization (e.g., angular discretization in Equations 14–16) is not analyzed.

Suggestions:

1.      Compare PCAG with Diffusion-based models in terms of generation quality and speed.

2. Conduct error analysis via statistical distribution of coordinate deviations.

4. Writing and Presentation

• Abbreviations (e.g., PCAG) are not fully defined at first mention.
• Section 2 redundantly describes cGAN improvements; condense technical details into a unified methodology section.
• Figures 6–9 lack scale bars and axis labels, reducing interpretability.

Suggestions:

Reorganize sections to separate methodology (Sections 2–4) and experiments (Sections 5–6).

5. Final Recommendation

This work presents a meaningful contribution to low-altitude point cloud generation. However, revisions are necessary to address:

1. Extended target types and benchmarking with cutting-edge methods;
2. Efficiency metrics and error analysis;
3. Writing clarity and figure annotations.

Comments on the Quality of English Language

• Abbreviations (e.g., PCAG) are not fully defined at first mention.
• Section 2 redundantly describes cGAN improvements; condense technical details into a unified methodology section.
• Figures 6–9 lack scale bars and axis labels, reducing interpretability.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 4 Report

Comments and Suggestions for Authors

1. The basic idea and processing flow of the article need a picture  to express.
2. Which formulas are proposed by the author for the need to be stated.
3. How the final experiment proves that the method can be promoted to the boundary conditions of actual application？
4. The discussion section is too long.
5. Some of the pictures are in color, some in black and white.

Comments on the Quality of English Language

 Too many long sentences in the article, for example,"The aircraft standard point cloud obtained in the manner described in Section 2.3.2 
 needs to be transformed into a depth image, which constitutes the standard data in Figure 1 The standard data, in combination with the channelized conditional infor-mation, forms  the real data utilized for training the improved cGAN. ".The entire paragraph is a sentence,hard to catch the point.

Author Response

Please see the attachment.

Author Response File: Author Response.docx