A Few-Shot Defect Detection Method for Transmission Lines Based on Meta-Attention and Feature Reconstruction
Round 1
Reviewer 1 Report
The paper proposes a few-shot defect detection method called "Meta PowerNet" for transmission lines based on meta-learning. It includes a new region proposal network (MA-RPN) and a meta-feature construction stage with a defect feature reconstruction module. The method achieves 72.5% detection accuracy for component defects with only 30 training objects. The MA-RPN designed in this paper can be used in other meta-learning object detection models universally.
The authors identify the problem of low detection precision of key component defects due to the scarcity of defect samples and propose using meta-learning to improve the generalization and fine-tuning of the detection model. They introduce the Meta PowerNet algorithm, which includes a meta-attention region proposal network and a defect feature reconstruction module. The paper concludes by stating that Meta PowerNet outperforms other few-shot object detection methods and can be extended to other meta-learning-based object detection methods.
However, the paper lacks a clear research question or hypothesis that the authors aim to answer with their proposed method. Additionally, the paper does not provide a detailed background or literature review of the problem of defect detection in transmission lines or few-shot learning methods. This makes it difficult to evaluate the novelty and contribution of the proposed approach. It seems the results can be further improved if the authors benefit from some geometric transformations as shown in Traffic congestion-aware graph-based vehicle rerouting framework from aerial imagery. So it might be necessary to discuss this paper as well.
Furthermore, the paper does not provide enough information about the experimental setup and dataset used to validate the proposed method. It is unclear how the authors evaluate the effectiveness of Meta PowerNet compared to other methods, and whether the results are statistically significant. In addition, as the paper Combined GANs and Classical Methods for Surface Defect Detection presents a method to improve the performance of surface defect detection by combining GANs with classical methods. The synthetic samples generated by the GANs are then used to train a deep convolutional neural network (CNN) to perform image segmentation and detect defects. Therefore, citing this paper can help to support the use of GANs to improve the performance of surface defect detection.
Regarding the future work, the paper only briefly mentions the issue of poor base-class detection performance and does not provide any concrete ideas for addressing this problem.
It might be better to add a paragraph indicating the organization of the paper to the of the Introduction section.
Usually, ablation study sections are placed before comparing the proposed method with existing studies (mostly the state-of-the-art). The ablation study evaluates the importance of individual components of a system or model. In the ablation study section, each component of the system or model is systematically removed and the resulting performance of the system or model is evaluated. This allows researchers to identify which components are most important for the system or model to achieve its desired performance.
The paper needs to provide more detailed information about the problem of defect detection in transmission lines, the dataset used, and the evaluation metrics. The authors also need to clearly state their research question and hypothesis and provide more ideas for future work.
The Fig. 5 must be improved or removed. An explanation paragraph is enough for those simple steps.
The figure size and bounding-box quality must be improved.
How many classes can the proposed method detect?
All of the figures seem a bit low-quality.
A discussion paragraph can be useful in the conclusions that evaluates the possible comparison with out-of-distribution models or can be provided as a future work.
The English used in the paper is generally good. There are some minor errors in grammar and syntax, but these do not significantly affect the readability or comprehension of the paper. Some sentences could be rephrased to improve clarity, such as in the abstract where it says "which has the ability to adapt to new tasks quickly, shows good performance in few-shot object detection and has good generalization in the face of new tasks." The phrase "in the face of new tasks" is somewhat confusing and could be rephrased for better clarity.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Overall the paper is well written, easy to understand. The application considered is interesting. In future, work may be expanded to improve accuracy beyond 72%.
There are minor grammatical errors which can be improved in the revised paper.
Author Response
Point 1: There are minor grammatical errors which can be improved in the revised paper.
Response 1: Thank you very much for giving us valuable advice. We have checked the paper again, and corrected some grammatical error.
Reviewer 3 Report
The manuscript is focusing on the recognition and identification of transmission line defects, applying a few-shot defect detection method (Meta PowerNet). The authors significantly extended the known algorithms with new modules which improved the accuracy of error detection. The tested components were insulators, vibration dampers and pins of transmission line. Commonly used evaluating methods were applied for characterizing the defect recognition performance of different objects. Based on this comparison the authors showed that the elaborated new method improved the accuracy of detection related to other solutions presented in other references.
The publication contains detailed summary of existing solutions showing their advantages and problems. It represents high level in the developed solution and application for Unmanned Aerial Vehicle inspection. The conclusions derived from the comparative results show that the reliability of the system reaches high level.
The manuscript contains small misprints mainly in citation of references (for example: Row 38, “Few-shot learning[5]”; Row 158, “a Siamese neural network [] as”) and minor English styling is necessary. Some words in figures are too small and do not have enough contrast, maybe these could be corrected.
Author Response
Point 1: The manuscript contains small misprints mainly in citation of references (for example: Row 38, “Few-shot learning[5]”; Row 158, “a Siamese neural network [] as”)
Response 1: Thanks for your suggestion, We have modified the error exist in our paper
Point 2: Minor English styling is necessary. Some words in figures are too small and do not have enough contrast, maybe these could be corrected.
Response 2: In order to show the figures and bounding-box clearly, we have enlarged the size of the figures, in addition, we change the format of some figures into MS visio for higher resolution. As for bounding-box quality, we enlarged the font size of labels beside bounding-boxes, we also croped the main parts of the detecion results for distinguishing easier.
Reviewer 4 Report
1. What is the main question addressed by the research?
The main question addressed by the research is how to improve the detection of defects in transmission lines using a few-shot defect detection method based on meta-attention and feature reconstruction. Specifically, the authors aim to address the challenge of detecting component defects in transmission lines with limited training data.
2. Do you consider the topic original or relevant in the field? Does it address a specific gap in the field?
The topic appears to be both original and relevant in the field. While there has been significant research on defect detection in transmission lines, few studies have focused on the use of few-shot learning for this task. This approach is particularly important for practical applications, as it enables the development of accurate defect detection models with limited training data. Thus, the research addresses a specific gap in the field by proposing a new method that improves the accuracy of defect detection with few training samples.
3. What does it add to the subject area compared with other published material?
The research adds a new method for few-shot defect detection in transmission lines, called Meta PowerNet. This method uses meta-attention RPN and Feature Reconstruction Module based on meta-learning, which improves upon other mainstream few-shot object detection methods by achieving 72.5% detection accuracy for component defects with only 30 training objects for various types of component defects. This is a significant improvement compared to previous studies on this topic.
4. What specific improvements should the authors consider regarding the methodology? What further controls should be considered?
The authors could consider exploring additional few-shot learning methods, such as Meta-SGD, to further improve the performance of their proposed method. Additionally, they could investigate the effect of different hyperparameters on the performance of the model, as well as perform a more thorough comparison of their method with other few-shot learning algorithms.
5. Are the conclusions consistent with the evidence and arguments presented and do they address the main question posed?
The conclusions appear to be consistent with the evidence and arguments presented in the paper, and they effectively address the main question posed. The authors demonstrate that their proposed method significantly improves the accuracy of defect detection in transmission lines with few training samples, and they provide evidence to support this claim.
6. Are the references appropriate?
The references cited in the paper appear to be appropriate and relevant to the research topic. However, it may be beneficial for the authors to consider citing additional studies on few-shot learning or defect detection in transmission lines to provide more context for their research.
7. Please include any additional comments on the tables and figures.
The figures in the paper are all low resolution and low quality, which makes it difficult to discern the details of the proposed method. It would be beneficial for the authors to improve the resolution and quality of these figures to enhance the clarity of their research findings.
I have no problem reading the English in this paper.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
The authors addressed all of my concerns successfully.
