Multiscale Feature Extraction U-Net for Infrared Dim- and Small-Target Detection

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The author extensively investigated infrared weak and small target detection in the manuscript. The following are some comments I have regarding the paper:

1. I recommend revising the contribution section of the paper to explicitly enumerate the innovative aspects, facilitating a better understanding of the author's contributions.

2. Numerous terms lack comprehensive explanations, such as SE-Net, ECA-Net, and CBAM. It is essential to provide detailed explanations of these terms in the manuscript.

3. Although the author utilized multiple open-source datasets in the paper, no new dataset was released. Consequently, I believe a dedicated subsection introducing the dataset is unnecessary.

4. While the author employed formulas to describe the computation processes in various modules, specific meanings of symbols in each formula, e.g., (1), (2), (3), were not provided. Detailed explanations of symbols need to be incorporated.

5. The author's calculations of metrics appear ambiguous, especially regarding the Probability of Detection (Pd) and the True Positive Rate (TPR). If these definitions are identical, it seems challenging to correlate the ROC curve mentioned by the author with Table 4. The definitions of TF and FP also need clarification.

6. The syntax and writing in the manuscript should be strengthened.

7. The author lacks relevant analysis of experimental results, including ablation experiments and comparative studies.

 

8. It is noted that the author did not introduce a new loss function. Therefore, it is advised to limit the extensive discussion on loss functions.

 

9. Has the author conducted detailed experiments on deep learning algorithms? Particularly, the results of the DNA algorithm on the IRSTD-1k dataset appear inconsistent with those in existing literature. Clarification on the number of experiments conducted is needed.

Comments on the Quality of English Language

The syntax and writing in the manuscript should be strengthened.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The paper presents a system for infrared and small target detection based on a multiscale feature extraction U-NET. The deep learning architecture incorporates several well-known principles: encoder-decoder architecture, residual, attention mechanism, multiscale information, Focal and Dice losses. In this sense, the novelty of the paper is average: there is not a very new deep learning approach. Despite this, the application to the specific domain of small target detection in IR imaging is of interest, the design choices provide valuable insights for practitioners and researchers, the extensive experimental evaluation and the comparison with several state-of-the-art algorithms is a merit, indeed. 

The major concern that I have is about the experimental design (Section 3.2): for the three available (and tested) datasets, the authors trained 3 different networks (because different images/input sizes are chosen, as well as different training hyper-parameters), and then tested them over the same dataset (after splitting in training and test). This choice is not optimal in my opinion: first, which one of the models should a user employ? Then, and mostly important: there might be the risk that each network overfits with respect to the specific dataset (MFIRST, SIRST, IRSTD-1k). In other words, generalization of the proposed architecture with respect to the input data might not to be guaranteed/demonstrated. 

Minor comments:

A careful review of the language should be done before publication. 

Check References:

- Anticipate reference [19] into the paragraph at lines [48-52].

- Reference [20] is missing. See lines [84-87]. Please chek it.

Comments on the Quality of English Language

A careful review of the language should be done before publication. Some examples are:

- The use of future (tens) in Section 1 is redundant and not needed.

- Lines [36-37]: “Detection …. Detection, application …. application, value … value” is full of unnecessary redundancy. 

- “network-based network” in line 70

- “Maxpooling” (line 204) vs “MaxPooling” (line 285): please, be consistent with the terms. 

- And others within the text

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

In this work a new infrared dim and small target detection network: Multi-scale 4 Feature Extraction U-Net for Infrared Dim and Small Target Detection (MFEU-Net), which can 5 accurately detect targets in complex backgrounds is presented. In general, the manuscript is well written and the results are convincing. Here some issues that encourage the authors to address:

 1.     There are a little grammatical/style error. In my opinion, a grammar/style revision has to be carried out before the manuscript can be considered for publication.

Comments on the Quality of English Language

 Minor editing of English language required

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

Comments and Suggestions for Authors

This paper proposes a new network called MFEU-Net for accurately detecting infrared dim and small targets in complex backgrounds. The algorithm incorporates a multi-dimensional channel and spatial attention mechanism to focus more on the target area in the image, improving target information extraction and detection performance in different scenarios. I suggest that this paper requires major revisions before publication. Specific suggestions for modifications are as follows:

 1 The abstract notes that the algorithm in this paper has the highest detection rates and IoU. It is recommended to provide specific numerical values in Abstract.

 2 The introduction should delve more deeply into analyzing the shortcomings of existing IDST detection neural networks, rather than just discussing the issues with datasets problems.

 3 “Related work”should not merely introduce others' work, but also highlight their limitations or the inspiration they provide for this study.

 4 The description of the method in Section 2.2 should be clearer and more concise. For instance, RSU and MCSAM should be directly indicated in figure 1 and their functions briefly explained in Section 2.2.1.

 5 I suggest that the author beautify Figure 2 and remove the dashed lines.

 6 In the article, variables and symbols should be explained under each equation.

 7 The captions under Figure 5 and Figure 6 have been mistakenly written the same, and it is suggested to correct this error.

 

Comments on the Quality of English Language

 In 'Comparision to the State-of-the-Art Method', 'Comparision' should be corrected to 'Comparison'. I suggest that the author carefully review the language expression in other parts of the article.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

All my concerns have been revised. The organization and clarity of the paper are good enough for publication. 

Comments on the Quality of English Language

The quality of English is good enough for publication.

Reviewer 4 Report

Comments and Suggestions for Authors

The author of this paper completed the revision according to the suggestions, and the quality of the paper has been improved