一种多目标自适应优化的伪装目标分割轻量化方法和系统
By constructing a camouflaged target segmentation model that includes boundary-guided and semantic-guided branches, and by optimizing the network structure using an improved genetic algorithm and a delayed lookup table, the problem of the imbalance between accuracy and efficiency in the lightweighting process of existing camouflaged target segmentation models is solved, and efficient camouflaged target segmentation is achieved on resource-constrained devices.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CENT SOUTH UNIV
- Filing Date
- 2026-04-27
- Publication Date
- 2026-07-17
AI Technical Summary
Existing COS models struggle to balance camouflage target segmentation accuracy and inference efficiency during the lightweighting process, making them difficult to deploy directly on resource-constrained embedded edge devices.
A camouflaged target segmentation model is constructed, which includes a fixed stage, multiple searchable stages, a boundary-guided branch, a semantic-guided branch, and a boundary-semantic gating fusion module. The network structure is optimized by an improved third-generation non-dominated sorting genetic algorithm. Combined with delayed lookup table and supernet training, a camouflaged target segmentation model with shared weights is constructed.
It improves the accuracy and inference efficiency of camouflaged target segmentation, and can be effectively deployed on resource-constrained embedded edge devices to achieve high-quality camouflaged target segmentation.
Smart Images
Figure CN122090072B_ABST