一种基于神经核方法与增量式深度网络的多模态图像融合方法
By employing a multimodal image fusion method based on neural kernels and incremental deep networks, high-quality pseudo-near-infrared features are generated from visible light images, solving the target perception challenge of spacecraft in low-exposure and high-noise environments and achieving efficient orbital threat identification and detection.
CN121526892BActive Publication Date: 2026-07-17HARBIN INST OF TECH
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HARBIN INST OF TECH
- Filing Date
- 2025-12-24
- Publication Date
- 2026-07-17
Smart Images
Figure CN121526892B_ABST
Abstract
本发明属于空间目标感知与探测技术领域，具体涉及一种基于神经核方法与增量式深度网络的多模态图像融合方法，包括步骤一：收集近地轨道环境中航天器目标的成对可见光VIS与近红外NIR多模态感知时序图像数据。步骤二：构建并预训练一个双模态可见光VIS+近红外NIR输入的教师融合网络模型；步骤三：搭建一个仅接收单模态可见光VIS输入的轻量级学生网络模型，步骤四：构建一个多层次知识蒸馏复合损失函数；步骤五：冻结教师模型参数。训练完成的学生模型仅需可见光输入，即可生成高保真度的融合特征图像，用于后续的目标检测任务，本发明利用神经核函数与余弦相似度结合，针对特征提取层获得更好的保持特征表示，更直观地反映两个模型的特征空间是否对。
Need to check novelty before this filing date? Find Prior Art