一种特征提取与特征交互增强的扣件螺栓检测方法
By introducing FRepBlock and FADDet into the track fastener bolt detection model, the problems of insufficient feature extraction and localization performance of existing models are solved, achieving high-precision detection and localization in complex environments and enhancing the model's adaptability and anti-interference ability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HUAHAI ENG CO LTD OF CREC SHANGHAI
- Filing Date
- 2026-05-11
- Publication Date
- 2026-07-17
AI Technical Summary
Existing track fastener bolt detection models need improvement in feature extraction and positioning performance, especially in complex track environments where high-precision detection and positioning are difficult to achieve.
The model's ability to extract bolt features is enhanced by employing the feature extraction module FRepBlock based on structural reparameterization technology and the asymmetric decoupled detection head FADDet. The detection scenarios are enriched by offline data augmentation methods. A multi-branch parallel structure is designed to perform topological decoupling in the training and inference states. The channel shuffling mechanism is combined to improve the bounding box regression accuracy.
The model significantly improves its ability to detect and locate bolts with high precision, enabling accurate identification and positioning of fastener bolts under complex working conditions, and enhancing the model's generalization and robustness.
Smart Images
Figure CN122156919B_ABST