基于强化学习调整的层次概率水下图像增强模型构建方法
By using the HPUIE-RL model, combined with reinforcement learning and hierarchical probabilistic techniques, the uncertainty and adaptability issues of underwater image enhancement methods in different environments are solved, achieving high-quality image enhancement and good generalization ability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANGHAI OCEAN UNIV
- Filing Date
- 2023-12-29
- Publication Date
- 2026-07-17
AI Technical Summary
Existing underwater image enhancement methods suffer from uncertainty in enhancing images when no real, distortion-free images are available, and they are not adaptable to different underwater environments. Deep learning methods rely on the quality of underwater image datasets, and traditional methods are difficult to effectively evaluate image quality.
A hierarchical probabilistic underwater image enhancement model (HPUIE-RL) based on reinforcement learning is adopted. Through multi-layer probabilistic image style transformation and multi-quality metric reward functions, combined with a prior encoder, deterministic decoder and feature-to-style distribution block (FTSD) module, the model is fine-tuned to improve adaptability and image enhancement performance.
It achieves the ability to enhance high-quality images in different underwater environments and has good generalization performance. It can effectively capture multi-scale features and complex interdependencies, and generate diverse and robust enhanced images.
Smart Images
Figure CN117809166B_ABST