Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

In this paper, the authors propose a  lightweight and novelty FMP-Net. In the proposed method, the Multiscale Frequency-Aware Modulation and Perturbation module and the Adaptive Wavelet Frequency-Aware Downsampling module  are considered in the two successive FMP stages. The writing is well, and paper's structure is good. My concerns are as follows.

1. Based on my reading, the authors' work is under the framework of YOLO series. Which is your based YOLO version?
2. From the proposed method framework in Figure, we can see that the most work focused on the improvement of Backbone? How about other part, such Nech and Head?
3. What are the differences between the FMP module and the existing works?
4. In experimental results, the performances of YOLO9s、YOLO9n、YOLO10s and YOLO11s should be considered.

Some minor revisions:

1. Please add explainations of the meanings of underline and bold format in the Table 1-4.
2. In Table 1, "Rcnn" should be "RCNN".
3. In Table 3, "CenterNe [48]t" should be "CenterNet [48]".

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This manuscript discusses a new ship detection method by introducing the mechanism of modulation and perturbation in the frequency domain based on a lightweight CNN detector. Specifically, the wavelet transform is firstly used to extract high-frequency features in different directions, and the feature expression is dynamically adjusted according to the global statistical information to realize the selective enhancement of the ship edge and detail information. Overall, the experimental results are comprehensive and convincible, showing the scientific soundness of this manuscript. However, it still requires improvements before its publication.

 

1. There are two ambiguities in the description of the AWFA module in Section 2.3: first, only Haar wavelets are selected to extract high-frequency components in horizontal, vertical, and diagonal directions, but no explanation is provided for why these three directions are optimal for SAR ship detection; second, the window size for wavelet transform is not specified, and the impact of different window sizes on the accuracy of high-frequency feature extraction is not verified, leading to doubts about the robustness of the module design.
2. There is missing information in the experimental datasets in Section 3.1: the SAR system parameters (such as radar band, imaging mode, etc.) of the four datasets are not introduced.
3. The comparative experiment in Section 3.3 fails to ensure that the comparative methods adopt the same training strategy as FMP-Net (e.g., it is not stated whether methods like Yolov8n and Faster-RCNN use the same data augmentation, optimizer parameters, and training epochs).
4. The ablation experiment in Section 3.4 has two problems: first, it does not design single-module ablation groups of "AWFA only", "variance modulation only", and "FGP only", and only conducts verification through the combinations of "AWFA + variance modulation" and "FGP", making it impossible to separate the independent performance gains of each module; second, it does not analyze the interaction effects between modules: for example, it does not verify whether there is a synergistic/antagonistic effect between AWFA and FGP, nor does it test the impact of the module addition order on performance, resulting in the inability to clarify the priority and combination logic of each module.
5. This paper discusses the SAR image processing, but the references are not up-to-date enough, and it is recommended to supplement them. For example:

1） Polarimetric SAR Ship Detection based on the Sub-look Decomposition Technology

2） PSO-based fine polarimetric decomposition for ship scattering characterization

Author Response

Please see the attachment.

Author Response File: Author Response.pdf