A Complex Background SAR Ship Target Detection Method Based on Fusion Tensor and Cross-Domain Adversarial Learning

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The topic, namely, ML or DL based ship detections from SAR imageries by regarding SAR pics as the same that of the optical ones, has been extensively discussed and studied in recent years. But they all share the same dilemma, that is, they are all difficult to explain from a physical perspective - EM scattering mechanisms, while are of low novelty from a neural network standpoint. So does this manuscript. 

The only highlight of this study is that the proposed method achieved higher precisions than  that of the original NN models, through modifying original NN structure or with additional optimization strategies. 

(i) The authors must explain the details more for why such a NN model was proposed? SAR image characteristics driven or something else, throughout this paper, it seems no distinct evidence can be found, just few words on potentials are not enough;

(ii) Why full polarization SAR images were selected? if the authors known the SAR imaging mechanisms well, then, it is deduced that VV, HH, VH/HV show totally different coherent patterns, notably under the sea scenarios. This leads a key problem, say,  does the authors mean the proposed method show superior performances over all the SAR images with full polarizations and is better than all other NN models simultaneously?  I don't think it is possible; if so, in comparison results, they at least should be presented regarding to the polarizations and different models respectively. 

Comments on the Quality of English Language

need moderate editing before next round submission 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Manuscript ID: remotesensing-3085014

 

In this paper, the authors have proposed a novel method to improve the performance and ship detection using the Channel Fusion Module (CFM) based on the YOLOV5s model that used the correlation between the polarised channels extracted during the inference.

The main idea of the paper seems to be in the scope of the journal. It is a valuable contribution to the scientific community, due to advancements in ship detection models. The paper is well organized, I suggest minor changes in the results section which should be more explicit and simpler. 

Comments:

1. Improvement of the English language is needed, especially in ABSTRACT and INTRODUCTION.
2. The last paragraph of the INTRODUCTION must describe explicitly (briefly) the main contribution added and the main difference with the state of the art. 
3. Moreover, the ABSTRACT need revision and should be more focused on the main contribution and proposed model instead of general applications and experimental results. 
4. Provide the reference of each equation used.
5. Can you describe what type of complex background you used? Is it a harsh condition of sea state or sea ROI that contains several multiple ships that create congestion?
6. In Figure 2, are you sure that images HV and VH are the correct ones? If it is amplitude images, they must be the same.
7. Can you add more details about the input to the model? Explain which polarization you provide at the input and why.
8. Can you explain the type of single polarisation that you used in the Conv+BN+SiLu structure or each single polarisation is tested one by one? It is still not clear that either you used a channel from multi-polarisation or you used each single polarisation. I will suggest rewriting section 3.3 to make the role of multi-polarisation more clearer.
9. Can you compare the detection speed of your methodology vs other target detector algorithms?
10. Can you provide the optimal size of the target? What should be the minimum/maximum distance of the target during detection?
11. Some relevant works on the ship's SAR imagery must be at least mentioned in the INTRODUCTION of the polSAR portion related to polarization and imaging parameters. See, just as a suggestion,

• On the Effects of the Incidence Angle on the L-Band Multi-Polarisation Scattering of a Small Ship. 2022.
• GK-based observation of metallic targets at sea in full-resolution SAR data: A multi-polarization study”, IEEE Journal of Oceanic Engineering, 2011.

 

 

 

Thanks

Comments on the Quality of English Language

described in main comments

Author Response

Please see the attachment

Author Response File: Author Response.pdf