Review Reports - An Improved YOLOv7 Model for Surface Damage Detection on Wind Turbine Blades Based on Low-Quality UAV Images

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This manuscript presents an AI-driven approach to detect surface damage to wind turbine blades. The authors utilize a modified YOLOv7 model to construct their algorithm. Three types of damage are being dectected, namely: cracks, delamination (known as "layers fall off"), and trachoma. The results presented appear sound, but some key concerns exist for this reviewer.

The general and most significant comment is the use of preidentified damage in Figures 5, 6, and 7 (as well as the results Fig 10-14). While it is noted that the images may be limited based on what is presented in this manuscript, it appears that a significant training bias exists that the damage has already been pre-identified. This can be related to data bias (sample and measurement biases).

This brings the question, how well does this algorithm work on images that are not preidentified. It is imperative that a system likes this work on the data that is not marked.

Other comments:
1. The use of "layers off damage" is very awkward. Check what is most common in this technical area. Perhaps this is delamination?
2. A training curve for the method would be very valuable demonstrating the MAP50 for all sets (training, verification, and test sets).
3. The conclusions are rather vague and lack specific details.

Comments on the Quality of English Language

No specific comments.

Author Response

Thank you for your valuable comments. Please review the attached file.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

1. As far as I know, there is an updated target detection algorithm. Why does the author still use yolov7? Please elaborate on the reasons. In addition, it can be seen from the comparative experiments in Table 3 and Table 4 that most parameters of other versions of YOLO are superior to YOLOv7.

2. In Section 3.2, we introduce the situation that the quality of some damage images increases the difficulty of damage detection, and whether these will affect the effectiveness of the training model.

3. The core of this paper is to put forward an algorithm, but whether adding an attention mechanism and some improvements to the YOLO model are not innovative enough, it is suggested to summarize again.

Author Response

Thank you for your valuable comments. Please review the attached file.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Hi,

To improve the accuracy and robustness of surface damage detection on wind turbine blades using UAV images of low quality, this paper proposes an enhanced YOLOv7 model. The enhancements include the integration of the ECA attention mechanism module, a down-sampling module, the DGST multi-scale feature fusion module, and the MIoU loss function module. The performance of this improved YOLOv7 model is compared with state of the art models.

However, I have the following concerns:

1. The dataset is private so the efficiency of the model is unclear. There are open-source datasets available to further check the efficiency of the model e.g. "Drone Footage Wind Turbine Surface Damage Detection",IEEE IVMSP 2022

2. There are terms not being defined in the manuscript such as CBR ,CONVFFN. Please define all the terms throughout the paper.

3. There are some recent papers missing from literature such as :

Y. Limei Ma, X. Jiang, Z. Tang, S. Zhi and T. Wang, "Wind Turbine Blade Defect Detection Algorithm Based on Lightweight MES-YOLOv8n," in IEEE Sensors Journal, doi: 10.1109/JSEN.2024.3430351. keywords: {Wind turbines;Feature extraction;Blades;Accuracy;YOLO;Sensors;Defect detection;Deep learning;defect detection;wind turbine blade;YOLOv8n},

Please add these and what makes your study different.

4. The modules mentioned used in different architectures e.g. ECA channel use in the neck of YOLOv7 ,MIOU ,down-sampling . Could you please further explain the novelty of this work.