基于声呐操作员大脑识别声信号神经机制特性的水下目标特征提取与自动识别研究

Recognizing underwater targets is critical for surface vehicles and underwater vehicles. There are two major ways to recognize underwater targets: One is to extract features of acoustic signals and classify the features using computers, and the other is that sonar operators listen to acoustic signals for recognizing underwater targets. However, how a skilled sonar operator processes and analyzes the acoustic signals and then determines which type of the underwater target belongs to, is not widely or deeply researched. Therefore, in this project, the first research theme is to determine the auditory discrimination and sensitivity of skilled sonar operators. Then, with the acoustic signals of real underwater target as the stimuli, the ongoing EEG data of the skilled sonar operators are collected; in terms of tensor (multi-way data array) decomposition, the EEG components which are significantly correlated with the features of acoustic signals are extracted; this allows analyzing the interaction between the sonar operators and the acoustic signals, and finding which features of the acoustic signals the sonar operators depend on to recognize the underwater targets. Finally, the coupled tensor decomposition algorithms are developed for analyzing the multi-sets of ongoing EEG elicited by the multiple underwater target samples; consequently, the EEG features correlated with the features of acoustic signals are extracted for multiple underwater target samples; the EEG features and the features of acoustic signals are fused together for classification to recognize the underwater targets. The outcomes of the research assist to develop the theoretical foundation to design new technologies for underwater target recognition, and assist to select the candidates of sonar operators and to supervise the training of sonar operators.

水中目标识别是水面和水下航行器目前急需解决的重要关键技术。当前，基于信号处理技术提取目标特征的识别和声呐操作员听音识别是两种主要的水下目标识别方法，但熟练的声呐操作员的大脑如何加工所听到的水下目标声信号，进而识别出目标种类，这尚未得到深入研究。鉴于此，本课题先通过脑电图-事件相关电位技术，研究熟练的声呐操作员的听觉辨识力和对声音特质的敏感性；其次，使用真实水下目标声信号作为实验刺激，采集熟练的声呐操作员的连续脑电波，通过张量（多维度数组）分解提取出与声信号特征相关的脑电图成分，从而挖掘出大脑与声信号之间的交互作用和识别水下目标所依赖的声信号特征；最后，通过耦合张量分解，提取出熟练的声呐操作员识别多个类型水下目标样本的脑电图特征，并且将该脑电图特征与识别目标所用声学特征融合组成新的水下目标特征，进而识别目标。该工作为发展水下目标自动识别技术进一步夯实理论基础，为选拔和培养声呐操作员提供依据。

本项目以声呐操作员为研究对象，以连续脑电图和事件相关电位为脑工程技术手段，研究了声呐操作员听觉系统辨识能力和其在听音判型任务下大脑反应模式，发展了联合水声与脑电信号实现水下目标识别的新算法。.对声呐操作员的听觉辨识力检测，取得巨大工程突破，具体可分为：（1）设计失匹配负波实验范式可实现声呐操作员听觉辨识力检测；（2）基于脑电信号时频分析发现声呐操作员对声音更加敏感的属性。本研究内容成果为声呐操作员的选拔与培训提供重要理论依据，可进一步落实为声呐操作员选拔内容的评测项目。.对水下目标识别技术上取得了突破性进展：（1）利用水声信号的辐射噪声实现基于卷积神经网络实现端到端的水下目标自动识别；（2）利用声呐操作员听音判型过程中采集到的脑电信号实现水下目标识别。这为水下目标识别提供了新信息，开辟了新思路。