Infrared Small-Target Detection Using Multidirectional Local Difference Measure Weighted by Entropy
Round 1
Reviewer 1 Report
This paper proposed the multi-directional local difference measure weighted by entropy(MDLDE) infrared small target detection method. The detailed comments are listed as follows.
1. The proposed MDLDE combines MPCM [21] and AADCDD [26] with no evident innovation. In addition, according to the experimental results from Table 1 and Table 2, the efficiency does not have obvious advantages.
2. The detailed information of the four sequences should be arranged in a table.
3. The layout of Table 1 is not easy to compare performance; it is recommended to rearrange it.
4. The paper claims that it is possible to handle dark targets in different situations using a weighting coefficient E(x). However, in practical cases, the brightness of targets may vary with the environment. If the brightness and darkness of the target are not known in advance, how to choose the weight coefficient E(x)?
5. This paper's description of the whole process is unclear; only Figure 3 is for reference, but it is not entirely consistent with the text. It is recommended to add more description of the detailed steps according to the steps in Figure 3 and add an example for an explanation.
6. The experimental samples consist of different numbers of targets. Seq.3 consists of three targets, both bright and dark. It should be worth discussing the differences in performance compared to other samples. In addition, since most IR small target detection methods detect a single target, is the experimental result comparison of Seq.3 fair for other methods that cannot be applied to multiple targets?
7. Some variables in the text are not in italics. Ex: n and L on page 7.
8. Experimental samples need to be extended, especially those composed of dark objects, to highlight the method's advantages. And the STOTA method of the past one or two years should be added for comparison.
9. The writing style must be enhanced to address the proposed method's advantages explicitly.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
It is still a challenging task to detect small targets from infrared remote sensing images. According to the shortcomings of existing detection methods, this paper proposes multi-directional local difference measure weighted by entropy (MDLDE) to detect small targets in infrared images with chaotic background. Generally speaking, the logic of this paper is clear, the theory is solid, and the experimental process is rigorous. Some minor amendments are listed below.
1、In section 2.1, in order to solve the negative contrast value caused by dark target, the weighting coefficient EX is constructed, and the cumulative directional derivative Ki is calculated. The above formula is clear, but the reasons for doing so are not fully explained.
2、The description of Figure 2 in Section 2.1 is not detailed enough, so the meaning of coordinate axis should be indicated in the figure or in the article.
3、The NTFRA in section 3.3 can't enhance dim targets, and it seems that it can't be well reflected in fig. 5. In this section, it is also mentioned that when a dark target appears in the scene, except MPCM and MDLDE, other algorithms can't enhance the dark target area. This can't be well reflected in Figure 5, so you should add more examples to demonstrate your point of view.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
The authors have proposed a method based on multi-directional local measure weighted by entropy to detect various small targets in infrared spectrum. Following are the queries.
1. This manuscript has been motivated from the authors earlier works and also from Nie et al. 'An infrared small target detection method based on multiscale local homogeneity measure' 2018. Kindly highlight the key difference and original contribution.
2. Pg. 3..line 84...typo 'when'.
3. In Methodology...the eight convolution kernel are used from ref 21..the seed point of any algorithm and then adding entropy measure. I cannot see much novelty for creating the MDLDE saliency map.
4. The infrared detector specifications are missing.
5. The distance factor for any target size is a crucial role. Kindly add the range of the targets.
6. The complete description of the data is also missing in the manuscript.
7. The timing complexity of the method is also not given.
8. How this method can discriminate between a drone, bird and other small flying objects. Should be included in your experiments.
9. Why Renyi entropy is used and how the value of alpha is derived. Any empirical study.
10. As this work is mostly related to the implementation of the algorithm, the data and the code should be uploaded to repository for reproducibility.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
In the research they present an entropy-weighted multidirectional local difference based measure called MDLDE to detect small targets in infrared images with messy backgrounds.
General comments:
To make the results obtained more robust, the test data set should be increased.
How did you come to choose the algorithm parameters?
Show how the parameters of all the detection algorithms were set.
Show statistical graphs where you can see the dispersion of the results obtained by the detection algorithms.
Show the computational complexity of all algorithms.
Show that the results obtained are not only due to chance. Perform a statistical significance test on the data obtained by the algorithms.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
1. The authors have modified their manuscript by addressing the comments. The revised manuscript is improved compared to the former version.
2. Please check the Table 1 layout which seems to exceed the margin.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 3 Report
The authors have partially answered my previous questions. Most of the replies are not satisfactory and still not clear.
1. The novelty of the method is still not clear from the answers of 1&3.
2. The infrared sensor specification such as manufacturer, pixel pitch, sensor resolution, data format etc are missing.
3. Distance factor.....technically not sound.
4. Birds, drones and other flying objects are very much distinguishable in infrared images.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 4 Report
The authors have made the requested changes. I suggest accepting the research article for publication.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 3
Reviewer 3 Report
The previous queries are not addressed properly.
The basic fundamental science of infrared imaging, specifically small target detection lies in the sensor pixel pitch. Without the details of sensor, the method lacks quantitative analysis of the data. It only shows few relative performance which can drastically change for a different sensor.
Also the kernel used in the method is not new one.
The authors are requested to kindly share the technical details so that a clear understanding of the work is reporter in public domain.
