基于海平面约束与分层生成式注意力的可见光水面目标检测方法
By employing sea level constraints and a hierarchical generative attention method, this study addresses the issues of insufficient background interference suppression and missing target features due to dense occlusion in existing technologies, achieving efficient visible light water surface target detection and improving detection performance.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA WATERBORNE TRANSPORT RES INST
- Filing Date
- 2025-12-17
- Publication Date
- 2026-07-17
AI Technical Summary
Existing visible light water surface target detection schemes do not fully utilize the unique physical prior laws of water surface scenes, resulting in insufficient background interference suppression, easy introduction of false features by frequency domain transformation, loss of target features due to dense occlusion, high false detection rate of background, and poor target recognition performance in densely occluded scenes.
Sea level curves are located through scene semantic segmentation, water surface candidate areas and non-water surface exclusion areas are divided, adaptive Gaussian filtering is performed to suppress static ripple interference, inter-frame difference is used to extract dynamic target regions, a hierarchical generative attention architecture is constructed, and conditional generative adversarial network is used to complete the missing textures in overlapping areas to generate complete target feature maps, which are then matched with a dynamic anchor box generator.
It effectively suppresses background interference, reduces the impact of false features, improves the recognition of small target features and the completeness and accuracy of target recognition in densely occluded scenes, and enhances the detection performance in complex water surface scenes.
Smart Images
Figure CN121640183B_ABST