Proportional Similarity-Based Openmax Classifier for Open Set Recognition in SAR Images

Round 1

Reviewer 1 Report

In this paper, the problem of open set SAR target recognition is studied. The authors analyze the limitations of Openmax classifier and propose a proportional similarity-based Openmax classifier, which utilize similarity metric to enhance the robustness and recognition performance of SAR ATR model. Specifically, the reviewer has the following comments to this work:

 

（1）In algorithm 2 and 3, the formula of calculating the output score based on SAR image is given. However, the method proposed in section 3 does not explain how to distinguish known classes and unknown classes. If the authors use a threshold-based method, it is recommended to explain the reason for selecting such a threshold.

 

（2）It is suggested to add an overall framework of the proposed method in the manuscript to make it easier for readers to understand.

 

（3）In the experiment section, the confusion matrix separates the unknown class samples according to their true labels, and the recognition results are all divided into unknown classes, so the final recognition confusion matrix is not a standard square. Here, I suggest that the author record the samples of known class in the first 8 rows and first 8 columns of the confusion matrix, and the unknown class as an extra category located in the 9th row and 9th column of the confusion matrix. In this way, the discrimination ability between known class and unknown class can be seen clearly.

 

（4）In the introduction section, the author omitted some important references about SAR ATR based on deep learning.

 

1、 Mixed loss graph attention network for few-shot SAR target classification. IEEE Trans. on Geoscience and Remote Sensing. vol. 60, pp. 1-13, Feb. 2022. Art no. 5216613, doi: 10.1109/TGRS.2021.3124336.

2、SAR Target Classification Based on Integration of ASC Parts Model and Deep Learning Algorithm. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol 14. pp. 10213-10225, 2021, doi: 10.1109/JSTARS.2021.3116979

3、Attribute-Guided Multi-Scale Prototypical Network for Few-Shot SAR Target Classification. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing. vol 14. pp. 12224 - 12245, 2021, doi: 10.1109/JSTARS.2021.3126688

Author Response

Please find the attached file

Author Response File: Author Response.pdf

Reviewer 2 Report

Very interesting paper, well written. My recomendation is to publish this paper as it is.

Author Response

The authors would like to thank the reviewer for the time spent on our manuscript.

Reviewer 3 Report

In my opinion, the paper has been written properly, which can address the goals. It is also well organized. All sections have explained the adequate information and materials to provide a scientific journal paper regarding presenting an Openmax classifier based on similarities among train and test samples.

I think the paper can proceed to publish after minor comments and suggestions.

Minor Comments:

• Abstract, Lines 9-10: Please rewrite this sentence. There are many prepositions, making it hard for readers to follow them.
• Lines 29-32: It needs a reference. 
• Equation 3: I did not know why the authors used the absolute value operator. The number of classes should be positive values. If I have not missed any point, please remove the absolute value operator.
• Lines 307-314: Sections 4.A, 5B, ...? 
• Line 419: should read Equation (4)?
• Lines 514-515: Please rewrite the sentence. What do "no threshold" and "no a priori information" mean?
• There is no consistency in the placement of figures and tables. Sometimes they are after the first citation and sometimes before. It is better to place them after the paragraph they were mentioned in the text for the first time.
• I suggest resizing some figures like Figures 4 and 13. Two or three times smaller can also conduct the idea. Also, I propose merging some figures into one or two figures as sub-figures, such as Figures 6-12, with a consistent size of Figure 14.

 

Author Response

Please find the attached file.

Author Response File: Author Response.pdf