HSF-DETR: A Special Vehicle Detection Algorithm Based on Hypergraph Spatial Features and Bipolar Attention.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Special vehicle detection is crucial for intelligent surveillance, emergency rescue, and reconnaissance.
• Complex environments pose significant challenges to the accuracy and robustness of current detection algorithms.

Purpose of the Study:

• To propose HSF-DETR (Hypergraph Spatial Feature DETR), an advanced algorithm for robust special vehicle detection.
• To enhance detection performance in complex and diverse environmental conditions.

Main Methods:

• Developed HSF-DETR integrating four modules: CSFNet backbone with CECG, HyperSFM network, DDFE with BEA and FEFFN, and SCFUB.
• Utilized hypergraph structures for high-order feature correlations and adaptive multi-scale fusion.
• Employed hybrid state-space modeling and dual-domain feature encoding for precise feature allocation.

Main Results:

• HSF-DETR achieved mAP50 of 96.6% and mAP50-95 of 70.6% on a custom dataset, surpassing RT-DETR by 3.1% and 4.6%.
• The model maintains computational efficiency with 59.7 GFLOPs and 18.07 M parameters.
• Cross-domain validation on VisDrone2019 and BDD100K datasets demonstrated strong generalization and robustness.

Conclusions:

• HSF-DETR offers a highly effective solution for special vehicle detection in complex environments.
• The proposed method shows significant improvements in accuracy and robustness, validating its practical applicability.