认知行为信号处理与模式识别

傅山　杨政　黄丹

摘要：本研究报告以驾驶舱设计为背景和出发点，通过对驾驶舱设计理念和布局规则的深入理解，结合人因学理论以及特定的飞行操控，将飞行员的肌电信号特征作为研究对象，通过信号的分解、相关性分析、选择、组合四个过程，选择合适的信号处理和模式识别的方法，从信息学角度揭示飞行操控中肌电信号的状态特征，并结合生理学、人因学、飞行器设计等理论得出飞行员的操作绩效，使飞机驾驶舱内的仪器仪表、操纵驾驶杆等合理有效地放入驾驶舱且满足飞行员的要求，形成一系列行之有效的信号处理体系，最终为驾驶舱的设计提供指导或参考，以及为驾驶舱适航符合性验证提供帮助。 研究过程通过搜集前人的肌电信号分析方法，在传统的傅里叶变换，时域指标和频域指标，小波变换等等方法的运用，发现这些方法在静态疲劳检测方面有很好的结果，但是运用在动态疲劳检测中效果不佳。随着希尔伯特黄变换的提出，EMD在生物信号处理、结构检测等非稳态、非线性信号上有很好的运用。本研究比较了EMD与EEMD在肌电信号分解中的性能， 提出基于EEMD 和Hilbert 变换的动态疲劳评价方法。实验证明基于平均瞬时频率的疲劳指标很好的表征动态肌电信号的疲劳趋势。 实验结果显示，我们提出的动态肌电信号疲劳特征指标（瞬时平均频率），可以监测飞行员生理疲劳参数的实时状态。并且基于现有的信号处理体系信号分解——相关性分析——分量筛选——分量重构，在揭示肌电信号物理意义并将其运用在理论研究和工程实践中都十分适合，为解决高维、多类、大量的数据（包括生物信号、飞行数据等）的采集，并进行信号分解后，结合相关性分析，提取出其中有意义、需要重点研究的分量，进行信号重组突出研究。

关键词：肌电信号；经验模态分解；希尔伯特黄变换

Abstract：We take pilots EMG signal as the research object to study cockpit design and layout rules combining theory and specific flight control. This signal goes through the decomposition， correlation analysis， selection， combination， selecting the appropriate signal processing methods and pattern recognition methods to reveal the status of EMG features in the flight controls. From the perspective of information science， combined with physiology， human factors， aircraft design， the performance of the pilots operating can be got to make the aircraft cockpit reasonable and effective. In addition， we will establish a standard signal processing system， and ultimately provide guidance for the design of the cockpit. According to the collected resource， the traditional methods of EMG analysis are the fast Fourier transform， the time domain and frequency domain indexes， wavelet transform， etc. We find that these methods are good at static fatigue， but do well in dynamic fatigue. As the Hilbert Huang transform was proposed in 1998， it has been widely used in the unsteady and nonlinear signal processing， such as biological signal and structure damage detection. We compare the advantages and disadvantages of EMD and EEMD in the decomposition of EMG and propose a dynamic fatigue evaluation approach on the basis of EEMD and Hilbert transform. Experiments show that the mean instantaneous frequency as the fatigue index has a good performance in the EMG dynamic fatigue. Our research proposes the index of dynamic EMG fatigue-mean instantaneous frequency. It can monitor the pilots muscle fatigue during the flight. And a signal processing system， including decomposition， correlation analysis， selection and reconstruction， has proved to be good at the EMG signal. The signal processing system based on the existing signals can provide help for the whole team， in order to reveal the importance and physical meaning of high-dimensional， multi-class， large amounts of data （including biological signals， flight data， etc.）. After acquisition and signal decomposition， combined with correlation analysis， the system can point out which signals are meaningful， needed to focus on.

Keywords：EMG；Empirical Mode Decompostion； Hilbert-Huang Transform