基于内嵌专业知识和经验深度学习模型的浅海声源识别方法研究

Research on Passive Recognition Method of Ocean AcousticSource in A Shallow Water Based on Deep Learning ModelEmbedding Knowledge and Experience

以突破传统的水下声源识别方法为研究目的,结合基于海洋声学环境理论和知识引导学习方法,建立基于内嵌专业知识和经验的深度学习模型,提高分类精度,增强识别的容错能力.从数值仿真方面,经声源参数化模型、水声传播模型和信号传递函数方法生成浅海环境中声学仿真数据,利用仿真数据训练一维深度神经网络;从实验验证方面,使用实验声学数据,对短距离内浅海声源信号分类,同时,对模型在海洋环境中的适用条件以及声学环境参数不确定性对模型性能的影响开展研究,解决识别模型应用于实际海洋环境背景下适应性差的难题.

Recent studies have demonstrated that ocean acoustic target recognition model can be established based on deep learning model embedding knowledge and experience.Insights from knowledge-guiding learning,this program analyzes the feature extraction of training samples and the domain knowledge of the ocean acoustical environment,and a model called ocean acoustic target recognition model of ocean acoustic target based on deep learning model embedding knowledge and experience is established.This research improves the recognition rate of the ocean acoustic target and solve the problem of poor environmental adaptability of recognition model by the actual ocean acoustic environment.Subsequently,through numerical simulation and experimental data analysis,the adaptability of this method in range dependent the scenarios with limited training samples and the effect of environmental uncertainty are researched.The study of this program is expected to substantially improve the environmental adaptability of the recognition model.