一种基于时空信息背景重构的红外小目标检测方法及系统
By constructing a spatiotemporal background reconstruction network, the background image is reconstructed using the spatiotemporal information of continuous infrared image frames and residual calculation is performed. This solves the problem of background modeling and target detection in infrared small target detection, and achieves stable detection and efficient use of labeled data under complex backgrounds.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ZHEJIANG UNIV
- Filing Date
- 2026-05-13
- Publication Date
- 2026-07-17
AI Technical Summary
Existing infrared small target detection methods struggle to fully utilize the spatiotemporal correlation between consecutive frames in complex backgrounds, rely on a large amount of pixel-level labeled data, and infrared small targets are easily weakened or lost during multiple convolution processes, resulting in limited detection performance.
A spatiotemporal background reconstruction method is adopted. By constructing a spatiotemporal background reconstruction network consisting of a multi-scale three-dimensional spatial feature extraction module, a spatially deformable three-dimensional residual enhancement module, a temporal aggregation module, and a two-dimensional reconstruction module, a three-dimensional spatiotemporal tensor is constructed using continuous infrared image frames. Unsupervised training is performed to reconstruct the background image and perform residual calculations to extract the target response.
It improves the stability of background modeling and the reliability of weak target detection in complex scenarios, reduces the dependence on labeled data, and enhances the applicability and detection accuracy of the method under different complex backgrounds.
Smart Images
Figure CN122415995A_ABST