HFC-YOLO11: A Lightweight Model for the Accurate Recognition of Tiny Remote Sensing Targets

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

1. Reproducibility and Code Availability

• Comment : The paper lacks explicit links to code, trained models, or implementation details (e.g., hyperparameters, training protocols). While datasets (AI-TOD, VisDrone) are cited, reproducibility would benefit from open-source repositories.
• Suggestion :
  • Share code and models on platforms like GitHub or Zenodo with a DOI.
  • Include training details (e.g., learning rate schedules, augmentation strategies, hardware specs) in the manuscript or supplementary materials.

 

2. Technical Clarity and Accessibility

• Comment : The EIOU loss formulation (Equations 9–10) is mathematically dense and may be inaccessible to non-experts. Similarly, the GhostBottleneck design assumes familiarity with GhostNet.
• Suggestion :
  • Simplify the EIOU explanation by breaking down the edge-sensitive penalty term and scale-adaptive weighting with intuitive examples or visualizations (e.g., a diagram comparing CIoU vs. EIOU).
  • Briefly contextualize GhostNet’s "cheap operations" in the related work to clarify why it suits shallow layers.

 

3. Depth of Limitations and Failure Cases

• Comment : The conclusion mentions limitations in handling "extreme scale variations" but does not elaborate on failure modes (e.g., severe occlusion, dense clustering).
• Suggestion :
  • Add a subsection in the discussion analyzing failure cases, supported by qualitative examples (e.g., false negatives in crowded scenes).
  • Propose targeted solutions for these limitations (e.g., attention mechanisms for occlusion, clustering-aware loss functions).

 

4. Terminology Consistency and Proofreading

• Comment : Inconsistent notation (e.g., "YOLOv11" vs. "YOLO11," "GhostBottleneck" vs. "GhostBottleneck") and minor typos (e.g., "YOLOv11n" in Table 2).
• Suggestion :
  • Standardize model names (use "YOLO11" consistently as per the paper’s title).
  • Proofread for grammatical errors (e.g., "Academic Editor" placeholder in headers, missing period after "Figure 7(d)").

 

5. Experimental Scope and Comparisons

• Comment : While comparisons with YOLO variants are thorough, the paper could strengthen its impact by:
  • Including lightweight models like NanoDet or EfficientDet-Lite for broader context.
  • Reporting inference speed (FPS) to emphasize edge deployment feasibility.
   
• Suggestion :
  • Add a comparative FPS analysis in Table 3, noting hardware specifications (e.g., "tested on NVIDIA Jetson Nano").
  • Briefly discuss trade-offs against NAS-based methods (e.g., EfficientDet) in the related work.

 

6. Practical Deployment Considerations

• Comment : The focus on lightweight design for UAVs/smart cities is well-motivated, but practical deployment challenges (e.g., power consumption, latency) are not quantified.
• Suggestion :
  • Include a brief analysis of computational efficiency metrics (e.g., energy consumption, latency on edge GPUs).
  • Discuss compatibility with embedded frameworks like TensorRT or OpenVINO.

 

7. References and Citations

• Comment : Most references are recent and relevant, but formatting inconsistencies exist (e.g., missing journal volume/issue in some citations).
• Suggestion :
  • Verify all references adhere to MDPI’s formatting guidelines (e.g., italicizing journal names, ensuring DOIs).
  • Cite recent works on lightweight detection (e.g., GhostNetV2, MobileNetV3) to contextualize GhostBottleneck’s design.

 

8. Visualization and Ablation Studies

• Comment : Figure 7 effectively demonstrates detection improvements, but ablation studies could better isolate module contributions (e.g., HRS vs. GhostBottleneck).
• Suggestion :
  • Add a bar chart in Figure 4 showing incremental mAP50 gains from each module.
  • Include feature map visualizations (e.g., P2 vs. P5) to illustrate HRS’s noise suppression.

Recommendtion : minor revision

Author Response

Thank you very much for your valuable comments and professional advice. Your suggestions have greatly contributed to enhancing the academic rigor of our article. In accordance with your recommendations, we have carefully revised the manuscript and made the necessary modifications.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

1- In the introduction, the authors provide sufficient background information about the subject of the study.

2- The authors use a lot of abbreviations in the introduction, but some abbreviations are not explained. Especially in the quoted parts, giving the explanations of the abbreviations in the first place where they are used will increase the comprehensibility of the manuscript.

3- The authors have adequately explained the studies in the literature on their topic by referring to the valid references in section 2.

4-The quality of Figure-1, 2 and 3' should be improved.

5- The proposed methodology is explained in sufficient detail.

6- The experimental envirement and experimental dataset used are described in detail.

7- Experimental results are shared and compared with the results of similar studies in the literature.

 

Author Response

Thank you very much for your valuable comments and professional advice. Your suggestions have greatly contributed to enhancing the academic rigor of our article. In accordance with your recommendations, we have carefully revised the manuscript and made the necessary modifications.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

I would like to thank the authors for submitting their manuscript entitled "HFC-YOLO11: A Lightweight Model for Accurate Recognition of Tiny Remote Sensing Targets". This is a lightweight model designed to improve the accuracy of recognition of tiny objects in remote sensing imagery by addressing problems such as resolution-semantic imbalance and inefficient feature fusion. The research details architectural modifications to the YOLO11s base model, including a reconstructed feature pyramid with a retained high-resolution P2 layer and the removal of the P5 layer.

The authors' idea is clearly explained and the experimental results are promising. 

Comments and suggestions to the authors
1 - To improve reproducibility, it would be interesting to share the source code of the proposed HFC-YOLO11 model as well as the scripts used for the experiments (e.g. in a repository such as GitHub).

2 - Consistently use a single name for the proposed loss function (EIOU). It would also be good to avoid variants such as "Extended", "Edge-Sensitive" or "Enhanced" in different parts of the text to avoid confusion. The same applies to references to modules such as C3k and C3k2 throughout the manuscript. 

3 - In Section 4.5 on visualisation analysis, the reference to Figure 4 should be corrected to refer to Figure 7 - Although the research is sound, the authors should consider adding a more explicit statement about the potential for significant replication of the findings and the concrete benefits that this replication would bring to the field of small object detection in remote sensing.

Author Response

Thank you very much for your valuable comments and professional advice. Your suggestions have greatly contributed to enhancing the academic rigor of our article. In accordance with your recommendations, we have carefully revised the manuscript and made the necessary modifications.

Author Response File: Author Response.docx