MGFNet: A Progressive Multi-Granularity Learning Strategy-Based Insulator Defect Recognition Algorithm for UAV Images
Round 1
Reviewer 1 Report
In this paper, the authors proposed a multi-granularity fusion network (MGFNet) to diagnose the health status of the insulator. According to the results of the comparison of MGFNet with state-of-the-art methods on accuracy, the proposed method exhibited excellent performance. Overall, the paper is well written, but if the following comments are taken into account and the paper is revised accordingly, I believe it will become a highly polished paper.
(1) Overall, there are typographical errors present. Please make sure to check for typographical errors related to this. For example, in the green section, there may be confusion as to whether 'mb' means Megabytes or Megabits.
(2) If there are any constraints or limitations to the proposed method, it is recommended to add such information to the revised manuscript.
(3) Consistency is needed in the symbols used. There are some that are in italics while others are not, so it is necessary to unify the terminology and symbols used throughout the paper.
(4) The dataset used and hyperparameters applied to the model training in a table format would improve readability as much as possible.
Minor editing of English language required.
Author Response
Please see the attachment
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The following issues need to be addressed：
1. The logic of the abstract section needs to be enhanced, such as using two ’furthermore’ consecutively at the end of the abstract.
2. What does the keyword 'power line inspection' refer to and its role in this manuscript, which is not reflected in the Abstract; In addition, the order of key words should be arranged according to their importance.
3. It is suggested that the organizational structure of this manuscript should be added in the introduction section.
4. The title of the manuscript mentions UAV images, but there is no clear relationship between them and the research object in the Abstract and conclusion.
English should be improved further.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
The authors propose a multi-granularity fusion network for insulator defect recognition with distortion correction and attention mechanism. The proposed method could be used in practical applications.
In my opinion, the authors should better illustrate the scale of the problem they are trying to solve through multi-granularity. For example, it may be a percentage of all damage cases - damage that cannot be detected with currently available methods.
In section 3.3. Progressive Multi-Granularity Learning Strategy the four steps are not explained clearly. The text should be improved
The authors should mention the exact dataset they used and provide the link for it
The authors should read the text carefully and correct the spelling and mistakes. Some of them are mentioned below:
The text should be checked proofreader
Fig. 4 last picture should be Thunderstroke
Fig. 5 description should be structure or architecture not struction
Line 223 I assume that: “Where [·] is the quotient of ? divided by ..” corresponds to the equation (2) ?
Table 1. should be Params
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
The article propose a multi-granularity fusion network (MGFNet) to diagnose the health status of the insulator. The work has some reference significance, but this paper mainly has the following problems:
1. As far as I know, there is a problem of not recognizing the whole insulator in the process of insulator identification, will the use of Traversal clipping module to cut it cause to aggravate this problem?
2. In line 211, ‘The reason for the distortion is that the network requires the input image to be square (i.e., have an aspect ratio of 1), while insulator images are typically rectangular with a large aspect ratio’. there is a problem of complex background interference in the transmission line scenario and that only insulators should not be entered in the input process.
3. A description of the two-stage is missing from the text.
4. As shown in Figure 11, the shapes and colors of the different types of insulators in the diagram are very different, and different defects are more distinct on different types of insulators; will this have a significant impact on defect identification if no differentiation training is carried out?
5. In table 5, a comparison of the two stages of the algorithm is missing.
The English of this article needs to be moderately revised.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
The authors have revised the comments and the manuscript can be accepted.
English language needs to be moderately edited .
