多尺度特征对比学习的红外目标跨频段图像分割方法
By employing an end-to-end training framework based on multi-scale feature contrastive learning, combined with target-aware contrastive learning and multi-level feature fusion, the problems of inconsistent feature representation and weak model generalization ability in cross-band segmentation of infrared images are solved, thereby improving the robustness of infrared target segmentation and the accuracy of small target segmentation, especially with excellent performance under conditions of few samples.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NORTHWESTERN POLYTECHNICAL UNIV
- Filing Date
- 2025-12-31
- Publication Date
- 2026-07-17
AI Technical Summary
Existing deep learning methods suffer from inconsistent feature representations, weak model generalization ability, and low accuracy in segmenting small targets and edges in infrared image cross-band transfer and segmentation tasks. They perform poorly, especially under conditions with few samples, and the contrastive learning framework fails to effectively utilize the physical correlation and repulsion of infrared cross-band samples.
An end-to-end training framework for multi-scale feature contrastive learning is constructed. Through target-aware contrastive learning and multi-level feature fusion, the discriminative feature extraction of infrared targets is enhanced in a self-supervised manner. Semantic information is integrated by utilizing a multi-scale fusion mechanism. Combined with few-shot training and data augmentation techniques, the segmentation performance of the model in cross-band scenarios is improved.
It significantly improves the robustness and generalization ability of infrared target segmentation, improves the segmentation accuracy of small targets and complex boundaries, and maintains superior segmentation performance, especially in extremely low-sample scenarios.
Smart Images
Figure CN121811042B_ABST