PFFNET: A Fast Progressive Feature Fusion Network for Detecting Drones in Infrared Images
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis article presents the design of PFFNET, a drone detection method for infrared images. The author discusses the structure and functioning of PFFNET in Section II and presents some results using an Open Source dataset. However, I believe that major revisions are necessary before accepting the paper. Here are the reasons for my suggestion:
1. Method Section: The description of the PFFNET lacks many important details and references. While the authors provide an outline of the network structure and explain some aspects in subsequent sections, there are very few citations to support their work. It raises doubts about whether the entire network was developed by the authors. I recommend that the reference parts and the authors' specific contributions be clarified in this section. Without this information, it becomes difficult to evaluate the novelty and contributions of their approach.
2. Results Section: The primary results presented in Figure 6 and Table 1 involve the performance of PFFNet-R and PFFNet-S, which were cited from [31] and [32], respectively. However, it is unclear what the results corresponding to the author's own method are. It is essential to include their results using PFFNET in Section II. Without these results, it is challenging to comment on the effectiveness of the methods employed.
3. Discussion Section: Surprisingly, there is no discussion provided in the paper. It is strongly recommended that the authors include a discussion section to analyze and interpret their results. This section is crucial for providing insights, addressing limitations, and discussing the implications of the findings.
4. Overall Presentation: The current presentation of the paper is inadequate for publication. The narrative of the paper seems to "borrow" a network, utilize public data, conduct some tests, and conclude without substantial depth. To improve the quality of the paper, it is important to provide a more comprehensive and coherent story, addressing the concerns mentioned above.
Here are some suggestions for improvement:
1. Title: The title suggests a focus on drone detection; however, it is not clearly conveyed in the paper. According to the paper itself, the network appears to work well with other small objects in infrared images as well. Consider revising the title to accurately reflect the scope of the paper.
2. Language and Grammar: Please thoroughly revise the English throughout the entire paper. There are grammatical and language issues that need to be addressed to enhance the readability and clarity of the content.
3. Abstract: When introducing abbreviations in the abstract, provide the full words at their first mention, such as "CNN" and "IoU." Additionally, the experimental results, including values like 2.53% and 81.03%, are not mentioned in the main body of the text. Please verify and ensure the consistency and relevance of these values.
4. Introduction: It is advisable to include references in the Introduction section. When discussing the problems addressed in the paper, acknowledge the contributions of other researchers who have worked on similar topics. Furthermore, reorganize the content in the introduction, such as dividing the long second paragraph on Page 2 into multiple parts since it covers two separate points. Additionally, revise the phrase "the infrared unmanned aerial vehicle target" on Line 93 as it is poorly worded.
5. Method and Results Sections: As previously mentioned, a complete revision of these sections is required. Provide a more detailed and comprehensive explanation of the method used. Additionally, clarify the results obtained by the authors' own method, as this information is currently absent.
6. Discussion: It is recommended to include a dedicated Discussion section to analyze and interpret the results obtained. This section will provide an opportunity to discuss the findings, address limitations, and provide insights into the implications of the study.
7. Image Quality: Revise the images in the paper, especially those with small sizes and poor resolutions, such as Fig. 7 and 8. Additionally, ensure proper formatting of figure citations in the main text, such as referring to them as Fig. 7 or Figure 8.
8. Data Availability Statement: Since the data used in the paper are sourced from a public dataset, the current Data Availability Statement may be incorrect. Please review and revise this statement accordingly.
Comments on the Quality of English LanguageThere is necessary to check the words, terms, and paragraphs, and revise the English.
Author Response
Please see the attachment
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for Authors1- Could you edit the captions of the figures and present it in a more proper way?
2- It would be imperative to enhance the quality of the figures in the paper. In the pdf version I have quality is barely acceptable.
3- Could you add to the discussion why results were much better with the SIRST Aug data set compared to the IRSTD 1k data set? Are your techniques data depend to a large extend?
4- Also please add why reductions in the case of (8,4) ratio where the best?
5- It will be good in tables 2,3, and 4 to add similar results for the IRSTD 1k data set (for the purpose of comparisons).
6- It is known that detection of IR images is easier than detection of RGB images. Why not to try the proposed methods for RGB images, at least as future work.
Comments on the Quality of English Language
English could be improved.
Further polishing and editing by English--first-language professional can help.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThis research method aims to develop a fast detection method for small infrared targets, addressing the issue of response value loss in the target area during the down-sampling process of UAV targets. While the method is innovative, there are some suggestions and concerns as follows:
1. The pixel count of infrared small targets in infrared images is extremely low, which makes them susceptible to interference signals during the feature extraction process. To mitigate this, it is recommended to employ LSM for quickly identifying local regions with visual saliency. Additionally, enhancing image features through thresholding is suggested prior to noise removal via filtering.
2. It is advised to carefully select the camera lens. Nowadays, many lenses offer a pixel resolution of 4K or higher.
3. The limitations of this study, such as the effective distance between the drone and the photographer, should be clearly stated.
4. Figure 7 should include a description indicating that the image has been filtered to establish the error value before and after filtering. Relevant literature references are recommended for further information.
“Using Drones for Thermal Imaging Photography and Building 3D Images to Analyze the Defects of Solar Modules”
“Robust Detection, Classification and Localization of Defects in Large Photovoltaic Plants Based on Unmanned Aerial Vehicles and Infrared Thermography.”
5. Figure 8 should be presented separately to improve readability.
6. In the Ablation Experiments, it is recommended to include quantitative indicators in addition to qualitative descriptions.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
Comments and Suggestions for AuthorsRegarding the first question addressed to the author, an adequate answer was not provided. I believe that image enhancement is a straightforward task and should be executed since it is the most important source of this research. It not only improves the quality of the article but also serves as the foundational work for this study.
On the other hand, the responses to the reviewer's questions should be incorporated into the revised manuscript with clear annotations for better clarity.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
