DPCSANet: Dual-Path Convolutional Self-Attention for Small Ship Detection in Optical Remote Sensing Images

Round 1

Reviewer 1 Report (Previous Reviewer 2)

Comments and Suggestions for Authors

The reviewer's comments have been addressed. Thank you. 

Author Response

Dear Reviewer, Thank you for your positive feedback and for acknowledging that our revisions have addressed your concerns. We truly appreciate your time and valuable input, which has helped us improve our manuscript. We look forward to any further guidance from the editorial team and hope our work meets the standards for publication. Thank you once again for your support. Best regards, Jiajie Chen

Reviewer 2 Report (Previous Reviewer 3)

Comments and Suggestions for Authors

Manuscript is acceptable is it’s current form. 

Author Response

Dear Reviewer, Thank you for your positive feedback and for acknowledging that our revisions have addressed your concerns. We truly appreciate your time and valuable input, which has helped us improve our manuscript. We look forward to any further guidance from the editorial team and hope our work meets the standards for publication. Thank you once again for your support. Best regards, Jiajie Chen

Reviewer 3 Report (New Reviewer)

Comments and Suggestions for Authors

When introducing the neural radiation field algorithm, the author can provide a more in-depth description of related concepts and add specific advantages compared to traditional methods. Moreover, it is possible to consider adding more evaluation indicators related to practical applications.

Comments on the Quality of English Language

The entire text should use unified proprietary vocabulary as much as possible, such as "signed distance function" and "SDF", which can be written in full for the first time and then unified for a better reading experience. Some words are more colloquial, yes. Academic emphasis on strong language description.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper proposed DPCSANet, a remote sensing small ship detection model. The proposed model consists of some key components termed dual-path convolutional self-attention module (DPCSA), high-dimensional hybrid spatial pyramid pooling module (HHSPP) and weighted focal regression loss (Focal CIoU). The proposed method has certain innovation and the experiments are relatively sufficient, having shown some merits for accuracy and model parameter (Params), as well as calculated burden (FLOPs). Moreover, this paper is also well-organized and easily followed. However, there still are some issues that should be concerned about.

 

1.       Current paper lacks the rationality that uses YOLOv5s as baseline.

2.       Although this paper shows the quantitative improvement by using proposed modules, it is suggested to show the attention maps for better verifying the effectiveness of proposed attention mechanism.

3.       Comparative methods are old. Please compare with more algorithms published in 2024 to support the state-of-the-art performance

4.       Detailing the software and hardware utilized in the experiments is essential for ensuring reproducibility. It is important to provide comprehensive information, e.g., the decay rate and decay epoch of learning rate.

5.       Current paper lacks the discussion of some recent works. Please make difference from them to clarify the novelty and merit. For example, as for DPCSA, some sparse self-attention module or gated attention module are proposed [1-2]. As for HHSPP, some similar pooling strategies have been proposed [3-4].

[1] Consistent Representation Mining for Multi-Drone Single Object Tracking. IEEE Transactions on Circuits and Systems for Video Technology, 2024.

6.       The authors are encouraged to publish relevant code and give a link in the abstract, which promotes the exchange of academic achievements and benefits to the development of whole community

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

The authors present an image processing model designed to extract both local textural and global contextual features from remote sensing image data for the monitoring and detection of small ships. While the study is promising, several areas require clarification and improvement to strengthen the paper's contribution and readability.

 

The abstract contains several undefined abbreviations, such as DPCANet, YOLOv5, and IoU. To improve readability, it is recommended to either spell out these terms upon their first use or minimize the use of abbreviations in the abstract altogether.

 

The paper heavily relies on references throughout, making it difficult to discern the specific contributions and novelty of the work. It would be helpful to clearly articulate the unique aspects of the proposed model and how it advances the state of the art in the field. A dedicated section summarizing the key contributions would greatly improve the paper's impact.

 

The proposed model appears to outperform the baseline under specific conditions, as indicated by improvements in metrics such as (R%) and (AP%). However, it is unclear whether these improvements fall within the margin of error.

Additionally, the model seems to underperform in other critical metrics, such as (FPS) and (P%). These discrepancies raise questions about the overall superiority of the proposed approach. Further analysis is needed to validate the model's performance across all relevant metrics.

 

The claim of the model's superiority would be more convincing if additional experiments were conducted under typical environmental conditions. For instance, evaluating the model's performance in scenarios such as fog, dust, or rough sea waves would provide a more comprehensive assessment of its robustness. It would also be beneficial to test the model under normal environmental conditions to establish a baseline for comparison.

 

The creation process of the DOTA-ship dataset is not clearly explained. Providing visual examples to illustrate the differences between the new and old datasets would greatly enhance the reader's understanding of the data used in the study.

 

Figure 5 lacks a color bar to explain the variation in color intensity. 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

1. (a) The description of the dual-path convolutional self-attention module (DPCSA) lacks sufficient detail. The authors should provide a more in-depth explanation of how this module integrates local and global features without introducing conflicts. 

2. (b) While the DPCSA and HHSPP modules are presented as novel contributions, their uniqueness compared to existing methods is not clearly demonstrated. A more thorough comparison with similar approaches in the literature is needed. 

3. (c) Consider including the following papers in the literature review to strengthen the context of the work: https://doi.org/10.1016/j.fraope.2024.100170, https://doi.org/10.1186/s12859-023-05462-2  

4. (d) The reported improvements in AP (0.9% to 11.4%) vary widely across datasets. A more detailed analysis of these variations and their potential causes would strengthen the paper. 

5. (e) The authors should discuss how well DPCSANet performs on other types of small objects in remote sensing images, not just ships. 

6. (f) Given the complexity of the proposed modules, a discussion on the computational cost (FLOPs/Prams) and inference time compared to the baseline model is necessary. 

7. (g) The paper lacks a comprehensive error analysis, particularly for cases where the model fails to detect small ships or produces false positives.

Author Response

Please see the attachment.

Author Response File: Author Response.docx