听觉感知在辐射源个体识别中的应用研究

The task of the radar specific emitter recognition is facing with new challenges of more complex electromagnetic environments. The intercepter needs to identify specific emitters under the conditions of complex noises, multipath interferences, the mixture of intentional modulation signals and the lack of signal information. In the meanwhile, when our auditory perception system facing the similar complicated condition, it can always catch the feature of the target speech and impel the decision made by our brain. Therefore, we try to firstly find the common points of both the auditory perception phenomenon and the electromagnetic environment cognition. Then we can apply the improved computational auditory scene analysis (CASA) to the frontend of the radar specific emitter recognition. Finally, the electromagnetic environment cognitive frontend of high performance is highly desired to provide distortionless enhanced signals and classification information for the subsequent recognition processing. This project aims on the crossing field of the auditory perception and the electromagnetic environment cognition. The research is conducted in the framework of time-frequency analysis similar to the spectrogram. The main content includes: (1) Auditory masking effect based emitter signal enhancement under the complex noise condition. (2) Auditory expectation phenomenon based emitter signal sparse representation and instantaneous frequency estimation. (3) Auditory subjective evaluation criterion based radar emitter threat signal perception, including intentional modulation recognition and the quantitative analysis of interferences.

辐射源个体识别面临复杂的电磁环境感知问题，截获方需要在噪声和多径干扰、有意调制信号混合及信号不完整等不利条件下，唯一地识别辐射源个体。而人类的听觉感知系统在类似的复杂环境中，往往能够较好地捕捉目标语音的信号特征，从而辅助大脑作出最终判决。因此，可以通过分析听觉感知和电磁环境感知的共性，建立或改进适用于辐射源个体识别的听觉感知模型，力求为后续识别提供无失真的增强信号及调制分类信息。本项目针对听觉感知和电磁环境感知的交叉领域，在语谱图类时频分析的框架下展开研究，主要内容包括：(1)复杂噪声下基于听觉掩蔽效应的辐射源信号增强；(2)基于听觉期望现象的辐射源信号时频稀疏重构和瞬时频率估计；(3)基于听觉主观评价准则的辐射源威胁信号感知，包括威胁信号有意调制识别和威胁度定量分析。

本项目针对复杂环境下的辐射源个体识别问题，重点结合听觉感知模型和机器学习算法，在信号预处理，特征提取和分类器设计方面展开研究。通过实测数据验证，项目完成的算法具有良好的特征增强和指纹识别性能，能够在复杂的非平稳环境下实现辐射源信号的可靠个体识别。项目具体完成的工作包括：（1）在信号预处理方面，解析推导了高斯白噪声在S变换域的分布特性。在此基础上，结合语音谱增强算法，提出新型S变换同时抑制高斯噪声和α平稳冲击噪声，从而实现复杂噪声环境下的信号特征增强；（2）在特征提取方面，重点在时频分析的框架下提取辐射源信号的非平稳特征，提出聚焦S变换对时频特征进行初步分辨和提取，在此基础上，利用时频率分布特征和多模糊函数近零切片特征进行融合，为后续分类器识别提供稳健的时频特征；（3）在分类器设计方面，重点在极限学习机和核机器学习的基础上，提出多隐层输出矩阵极限学习机，特权信息极限学习机和多核融合框架下的支持向量机，有效提高了辐射信号的分类识别精度；（4）在辐射源威胁信号分析方面，在时频领域利用腐蚀算法提出能量聚集性判定准则，能够对非平稳信号的时频聚集性进行定量判断。通过项目的研究，较好的解决了复杂环境下辐射源个体识别中的部分关键问题，为后续研究提供了较为可靠的基础。