| 基于知识引导学习模型的浅海声源识别方法及其算法研究 |
| Research on Passive Recognition Method of Ocean Acoustic Target Based on GuidingLearning Model |
| |
| DOI：10.3969/j.issn.1003-2029.2025.03.008 |
| 中文关键词: 声源识别 引导学习 声场 物理学知识 声学环境参数 |
| 英文关键词:underwater acoustic guiding learning sound field knowledge of physics acoustic environmental parameter |
| 基金项目:自然资源部海洋环境探测技术与应用重点实验室资助项目（MESTA-2022-B006）；天津市科技计划项目资助项目 （24ZYCGYS00680）； 自然资源部海洋观测技术重点实验室资助项目（2021klootB02，2021klootB01） |
|
| 摘要点击次数: 393 |
| 全文下载次数: 234 |
| 中文摘要: |
| 基于海洋环境知识和引导学习方法构建水下声源识别方法模型研究是近年来海洋声学 领域的研究前沿和热点 ，在水下声源识别方面具有重要应用价值 。本文建立了基于内嵌专业知 识和经验的深度学习模型的浅海声源识别方法 ，构建了水声环境空间中声场模型和物理学知识 模型 ，提出了模型优化方法和水下声源识别算法 ，构建基于以物理信息神经网络为内核网络的 深度学习模型 ，该模型提高了分类精度 ，增强了识别分类的容错能力； 基于模型在海洋环境中 的适用条件及声学环境参数不确定性 ，对模型性能的影响规律开展研究 ，提升了对声源追踪和 分类识别的性能 ， 以解决应用于实际海洋环境背景下识别模型环境适应性差的难题 。在此基础 上 ，开展了海洋声学环境中水下声源浅海声源被动声学识别算法实验验证 ， 以实现海洋声源较 高精度、较高分辨率识别分类。 |
| 英文摘要: |
| Recent advances in underwater acoustics have focused on developing source-identification models that fuse ocean-environment knowledge with guided-learning strategies, an area that holds considerable promise for practical applications. This study proposes a shallow-water acoustic-source identification framework built on a deep-learning model enriched with embedded domain expertise. By integrating a spatial sound field representation with knowledge of physics, a model-optimization scheme and a dedicated recognition algorithm are designed. The resulting architecture, which employs a physics-informed neural network as its core, boosts classification accuracy and improves fault tolerance in source identification. We further analyze how model performance varies under different marine operating conditions and in the presence of uncertainties in acoustic-environment parameters. These investigations enhance the system 爷s capabilities for source tracking and classification, addressing the limited environmental adaptability of conventional models. Finally, the passive acoustic recognition algorithm through shallow-water experiments in realistic ocean settings is validated, thereby demonstrating high-precision, high- resolution source classification. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
