Typical Fault Detection on Drone Images of Transmission Lines Based on Lightweight Structure and Feature-Balanced Network
Round 1
Reviewer 1 Report (Previous Reviewer 2)
Thank you to the authors for their effort in improving the manuscript.
All my concerns have been fulfilled.
Author Response
Please see the attachment
Author Response File: 
Author Response.pdf
Reviewer 2 Report (Previous Reviewer 3)
Fig. 14 All images should be in the sam page.
Fig. 17 Caption should be in the same page of the figure.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 3 Report (Previous Reviewer 4)
Aiming at the problem of low accuracy and limited computational resources caused by the complex and variable scale of various defects in the transmission line inspection process of UAVs, an improved algorithm based on YOLOv7-Tiny is proposed. According to the experimental results presented in the study, it is claimed that it compresses the number of parameters by 74.79% and increases the computational effort by 66.92% compared to the original YOLOv7-Tiny, and it is presented that it improves the mAP (average Average Precision). 0.71%. The Jetson Xavier NX device was used to simulate the UAV inspection process, and the results claim to run at 23.5 FPS.
My comments about the study are as follows:
The overall writing, layout and flow of the article is very professional.
In the proposed study, YOLO, an already proven architecture, was used. From this perspective, the proposed method is only at the level of hyperparameter optimization or minor improvement in architecture. Looking at the article from this perspective, I think it will probably be rejected from a SCI-E indexed journal with a deep learning or image processing scope due to novelty or insufficient contribution. The difference made by the small improvement here should be shown in the drone images.
Can the authors prove that they developed yolo and actually made a scientific contribution? It is quite easy to write numbers through tables. Personally, I would like to see how other models in the Yolo family compare in real-time videos. Authors must share proof content from a link on youtube, github etc.
When I look at paper through a drone scope, it seems hypothetical that the images were captured from the air and run on a processor that could run on a drone. So, even if these images are in the sea and we assume that Jetson Nano can work in an unmanned submarine, this paper also becomes a marine science study. Ultimately, the fact that such a system has not been implemented in the drone is a big question mark in terms of journal scope. Perhaps the authors should seriously consider presenting the information they obtained during the flight with video evidence. There are differences between a flying device and a device that operates on static images. It is important to present raw and real-time videos of the processed images as evidence in order to be of evidentiary value. Also code, model etc. Sharing the contents in terms of evidence will be reassuring for the editor and the referee. To fit/adapt it to the evidence and journal scope or to prove that it actually works in this field? The writers must decide that.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report (Previous Reviewer 4)
The authors' answers are sufficient. However, I think that some of the points they mentioned are still not emphasized enough in the text. Perhaps the solution would be for the authors to clearly write that they offer a method that provides better performance from the Yolo family, instead of "improved algorithm based on YOLOv7-Tiny".
In addition, they should present the proving videos they provide to us (including Github, if possible) as an acknowledgment at the end of the article.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
Round 1
Reviewer 1 Report
The "drone tayloring" is still weak, but it's a matter of choice of the editor, overall, although the scenario is not that much challenging, this work is interesting.
I'd suggest using shorter sentences sometimes.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 2 Report
The paper proposes a fault detection algorithm for transmission lines. The algorithm is based on a lightweight module and a feature-balanced network.
The aim of the proposed fault detection strategy is to improve the standard fault detection in the transmission line by:
1) Providing four target datasets, which include insulator, damaged insulator, shock hammer corrosion and birds nesting on transmission line;
2) using a ghost module that aims to generate more features using fewer parameters; and
3) Introducing scSE and PA-Net and using CIoU in the YOLOv7-Tiny network and the NWD loss function to solve the problem of low detection accuracy.
From an application point of view, I think the paper is interesting. However, the research background in relation to the concepts of fault detection and isolation (FDI) and fault tolerant (FT) algorithms still needs to be improved.
The importance of the FDI and FT algorithms in the context of engineering applications needs to be reinforced by the authors. In particular the concept of "false alarm" and "missed alarm" need to be discussed.
In view of this I suggest to improve the research background by discussing about FDI strategy. In this respect the authors can refers to the following references to provide a better literature review.
[ - ] Lunze, J. and Richter, J. H. (2008). “Reconfigurable fault-tolerant control: a tutorial introduction”. European Journal of Control, 14(5):359–386.
[ - ] Merrill, W. C., DeLaat, J. C., and Bruton, W. M. (1988). “Advanced detection, isolation, and accommodation of sensor failures-real-time evaluation”. Journal of Guidance, Control, and Dynamics, 11(6):517–526.
[ - ] Behzad, H., Casavola, A., Tedesco, F., Sadrnia, M. and Gagliardi, G., ”A Fault-Tolerant Sensor Reconciliation Scheme based on Self-Tuning LPV Observers”. In Proceedings of the 15th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2018) - Volume 1, pages 111-118, ISBN: 978-989-758-321-6, doi: 10.5220/0006840501110118
[-] Blanke, M., Kinnaert, M., Lunze, J., Staroswiecki, M.: Diagnosis and Fault-Tolerant Control. Springer, Berlin (2003)
[-] S.X. Ding, Model-Based Fault Diagnosis Techniques, Springer-Verlag London 2013, DOI https://doi.org/10.1007/978-1-4471-4799-2
[-] Frank, P.M., Ding, S.X., Marcu, T.: Model-based fault diagnosis in technical processes. Trans. Inst. Meas. Control 22, 57–101 (2000)
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 3 Report
The paper proposes a fault detection algorithm for transmission lines using drone images.
The aim of the proposed fault detection strategy is to improve the standard fault detection in the transmission line by providing four target datasets, using a ghost module and introducing scSE and PA-Net to solve the problem of low detection accuracy.
The paper sounds better after the author's update. still need some update as follow:
1/ I suggest adding the word “drone images” to the title, in order to be in line with the special issue, such that "Typical fault detection on Drone Images of transmission lines based on light-weight structure and Feature balance-network"
2/Please, indicate a reference for each equation mentioned in the paper. Except for the equations from your own.
3/ Please, indicate a reference for each Figure demonstrated in the paper. Except for the Figures from your own.
4/ All others notes are shown inside the yellow box on the paper.
Comments for author File: Comments.pdf
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 4 Report
The work is not reader friendly in terms of writing. First of all, what are the promised/claimed scientific contributions in this study should be explained in articles. The proposed method should then be presented in detail. But in the method proposed here, we only see different versions of YOLO at each step. Shared code content is also someone else's repository.
The study seems to consist only of a model proposal that can be applied on the drone. In other words, it consists of the use of ready-made code blocks that can run on a hardware that can be run on a drone. I do not see any scientific contribution in this respect.
In case the said hardware is carried on the drone, it is the edge computing level response that is important, that is, the frame rate of all the relevant models and the response time for 1 frame should be compared. However, I could not see them in the article content and tables. Apart from these, detailed information of the data set and accuracy values on the basis of class are required.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 4 Report
non-published material submitted in the submission is a file belonging to someone else. In this regard, it is essential that the necessary regulations and the necessary material are loaded properly.
I think this paper is at the level of an ordinary YOLO application, there is no scientific innovation, contribution or content that can be added to it. I think it's an over-published study topic now. As a subject, it is only adapted to the drone magazine scope. In my opinion, it is not appropriate to publish it in the SCI-E index.
