Quantifying Multi-Scale Performance of Geometric Features for Efficient Extraction of Insulators from Point Clouds

Round 1

Reviewer 1 Report

This article proposes a method for automatically extracting insulators from point clouds by quantifying multi-scale performance of geometric features, with good effectiveness and robustness. It is a meaningful work of insulator extraction.

But there are still some issues that need to be modified, as follows.

1.In the conclusion section, there is a sentence that is not clearly expressed. “The results given by the method tested on 82 different pylons are expected, with an F1-score of 99.12% for tension insulator extraction and an F1-score of 97.46%. ”(Page 17)

2. Entropy-weighting method (EWM) appears multiple times which in introduction and 3.2.3 Qualification of Multi-scale Feature.

3. Try to use reference instead of author name. for example, “ In the first step, an approach proposed by Zhang et .al is introduced to extract the pylons [14]” can be changed as “In the first step, an approach proposed by reference [14] is introduced to extract the pylons”.

4. In the discussion part, you should add comparisons with sufficient similar public researches/references, and give analysis or explanations why your method can function better.

Author Response

• In the conclusion section, there is a sentence that is not clearly expressed. “The results given by the method tested on 82 different pylons are expected, with an F1-score of 99.12% for tension insulator extraction and an F1-score of 97.46%. ”(Page 17)

Response:

In lines 538-540 the vague sentence has been revised by adding an accuracy description for suspension insulators.

 

• Entropy-weighting method (EWM) appears multiple times which in introduction and 3.2.3 Qualification of Multi-scale Feature.

Response:

In line 288, the “Entropy-weighting method” has been replaced with “EWM”, and the similar issues have also been addressed.

 

• Try to use reference instead of author name. for example, “ In the first step, an approach proposed by Zhang et .al is introduced to extract the pylons [14]” can be changed as “In the first step, an approach proposed by reference [14] is introduced to extract the pylons”.

Response:

In lines 218-219, the sentence and the similar issues have also been addressed.

 

• In the discussion part, you should add comparisons with sufficient similar public researches/references, and give analysis or explanations why your method can function better.

Response:

In lines 490-500, By fully analyzing the related works and combining their self-described limitations, some advantages of our method have been listed in Chapter 5.1. We are aiming to extract insulators from more complex scenarios rather than be limited by pylon shapes or insulator types.

Author Response File: Author Response.pdf

Reviewer 2 Report

This manuscript is very interesting as a paper for efficiently extracting insulators from point clouds. The methodology presented by the authors seems to have adequate accuracy for automatically extracting insulators from point clouds. The authors clearly explain the research methodology through the methodology, and materials are suitable to verify the research results. In addition, the research results provide reasonable evidence for the subject made by the authors.

 

However, there are some minor questions about the research logic.

1. 2p, lines 44-77, the authors introduce the advantages and disadvantages of extracting insulators using the LiDAR system. As noted by the authors, object extraction from point clouds is still manual and repetitive so far. On the other hand, with references [5, 8], the authors say that the LiDAR system is capable of high-precision terrain information and object coordinates. Can this result be seen as an advantage compared to conventional optical imaging? The up-to-the-minutes photogrammetry can extract a high-precision point cloud and measure the position of an object even with only optical images. It would be nice to briefly compare object extraction between using optical images and using LiDAR, and mention why the authors chose the point cloud from LiDAR rather than the point cloud from optical imaging as the subject of their study.

 

2. The results of the study seem to focus on the object identification accuracy and location accuracy. In 3p, line 117-120, the authors cite [8] and mention the improvement of efficiency as one of the expected outcomes of the study. Does the suggested method not require additional time and cost compared to the method discussed in [8]? If so, it makes sense.

 

Although it is a minor part, I think a little revision would make perfect.

Author Response

• 2p, lines 44-77, the authors introduce the advantages and disadvantages of extracting insulators using the LiDAR system. As noted by the authors, object extraction from point clouds is still manual and repetitive so far. On the other hand, with references [5, 8], the authors say that the LiDAR system is capable of high-precision terrain information and object coordinates. Can this result be seen as an advantage compared to conventional optical imaging? The up-to-the-minutes photogrammetry can extract a high-precision point cloud and measure the position of an object even with only optical images. It would be nice to briefly compare object extraction between using optical images and using LiDAR, and mention why the authors chose the point cloud from LiDAR rather than the point cloud from optical imaging as the subject of their study.

Response:

In lines 55 to 59, a brief comparison is made between LiDAR and photogrammetric methods. Unmanned aerial vehicle LiDAR (Light Detection and Ranging) is the primary method for point cloud acquisition in current power corridor inspections. Your comments inspired future work that the multimodal fusion could incorporate more information such as RGB to better extract insulators.

 

• The results of the study seem to focus on the object identification accuracy and location accuracy. In 3p, line 117-120, the authors cite [8] and mention the improvement of efficiency as one of the expected outcomes of the study. Does the suggested method not require additional time and cost compared to the method discussed in [8]? If so, it makes sense.

Response:

In lines 119-122, we apologize for any unclear expressions in our original text. The sentences have been revised. It was originally intended to express that insulator extraction from LiDAR data can help UAVs conduct safe route planning and obtain insulators’ images at close range, which can improve the accuracy and efficiency of fault detection based on images. Although the efficiency of insulator extraction is not currently considered, it generally meets the post-processing requirements of point clouds in practical applications thanks to the preprocess of voxel sampling. We briefly implemented the efficiency in line 442. Real-time extraction of insulators remains a topic worth researching in the future.

Author Response File: Author Response.pdf

Reviewer 3 Report

It is a study conducted in a persuasive way on a very interesting topic. Please check the following few things.

- Please add explanation about Figure 4. Also, since the y label of (a) can be mistaken for the y label of (b), please adjust the picture.

- Please align the Greek letters 260-262 into lines. After this, the Greek letters need to be sorted out.

- The graphs in Multi-scale feature extraction and comprehensive scores in Figure 3 cannot be found in the text. Did you remove it because it's not important? Readers may have questions

- No explanation for Figure 5.

- Please describe further your consideration of the poor case (Figure 8). Under what conditions is it difficult to distinguish? To prevent this, what technical method should be used in the point cloud acquisition step? Or what additional algorithms can be used?

- Finally, please check for typos.

Author Response

• Please add explanation about Figure 4. Also, since the y label of (a) can be mistaken for the y label of (b), please adjust the picture.

Response:

In lines 223-225, the explanation about Figure 4 has been added and the y labels of (a) have been moved from the right side to the left side.

 

• Please align the Greek letters 260-262 into lines. After this, the Greek letters need to be sorted out.

Response:

In lines 266-268, the sentences have been sorted out.

 

• The graphs in Multi-scale feature extraction and comprehensive scores in Figure 3 cannot be found in the text. Did you remove it because it's not important? Readers may have questions

Response:

In lines 205-206, the “multi-scale feature evaluation” has been added.

 

• No explanation for Figure 5.

Response:

In lines 308-309, and 313-314, the explanation for Figure 5 has been improved.

 

• Please describe further your consideration of the poor case (Figure 8). Under what conditions is it difficult to distinguish? To prevent this, what technical method should be used in the point cloud acquisition step? Or what additional algorithms can be used?

Response:

In lines 451-458, we added more description for Figure 8 and offered some ideas to avoid poor cases.

 

• Finally, please check for typos.

Response:

Some spelling errors have been revised like line 488.

Author Response File: Author Response.pdf