基于特征聚合的目标检测方法及系统
By constructing a target detection network based on feature aggregation, the problems of high computational load and insufficient detection accuracy of traditional networks on embedded chips are solved, achieving efficient and reliable target detection, especially performing well in large-scale target detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CENT SOUTH UNIV
- Filing Date
- 2026-04-09
- Publication Date
- 2026-07-17
AI Technical Summary
Traditional target detection schemes based on deep convolutional neural networks have high computational load and insufficient detection accuracy on embedded chips. Lightweight networks, on the other hand, lack the ability to macroscopically control large-scale targets and fuse multi-scale features, resulting in decreased detection accuracy and reliability.
A target detection network based on convolution, pooling, dual-channel, spatial projection, and channel projection schemes is constructed. Features are extracted through micro and macro branches, and target detection is achieved through feature aggregation, signal rectification, and target projection. A global descriptor generation layer and a gated fusion matrix are used to improve feature representation and detection accuracy.
While reducing computational load, it improves the reliability and accuracy of target detection, especially in multi-scale target detection, significantly improving detection accuracy and the fineness of cross-level feature redistribution.
Smart Images
Figure CN122115839B_ABST