端到端的生理信号片段目标检测方法
By constructing a backbone network and a cross-stage locally parallel Transformer module through an end-to-end physiological signal fragment target detection method, the problems of interference and inaccurate localization in OSA event detection are solved, and efficient localization and automated annotation of OSA events are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- XIAN UNIV OF TECH
- Filing Date
- 2023-11-09
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies suffer from significant interference from normal event signals in OSA event detection, leading to decreased detection performance. Furthermore, they cannot dynamically determine the start and end times of OSA events, limiting the accuracy of automated annotation.
An end-to-end physiological signal fragment target detection method is adopted. By constructing a backbone network layer and a cross-stage local parallel Transformer module, combined with feature extraction and position encoding, a one-dimensional target detection loss function is used for training to locate the start and end times of OSA events.
It improves the accuracy of OSA event detection and the accuracy of automated annotation, alleviates the data hunger problem of Transformer models and the receptive field limitation of CNN models, and achieves efficient localization of OSA events.
Smart Images
Figure CN117503059B_ABST