Perez Reynoso Francisco D, Niño Suarez Paola A, Aviles Sanchez Oscar F, Calva Yañez María B, Vega Alvarado Eduardo, Portilla Flores Edgar A
Instituto Politécnico Nacional, Escuela Superior de Ingeniería Mecánica y Eléctrica, Mexico City, Mexico.
Departamento de Ingeniería Mecatrónica, Universidad Militar Nueva Granada, Bogotá, Colombia.
Front Neurorobot. 2020 Sep 29;14:578834. doi: 10.3389/fnbot.2020.578834. eCollection 2020.
Although different physiological signals, such as electrooculography (EOG) have been widely used in the control of assistance systems for people with disabilities, customizing the signal classification system remains a challenge. In most interfaces, the user must adapt to the classification parameters, although ideally the systems must adapt to the user parameters. Therefore, in this work the use of a multilayer neural network (MNN) to model the EOG signal as a mathematical function is presented, which is optimized using genetic algorithms, in order to obtain the maximum and minimum amplitude threshold of the EOG signal of each person to calibrate the designed interface. The problem of the variation of the voltage threshold of the physiological signals is addressed by means of an intelligent calibration performed every 3 min; if an assistance system is not calibrated, it loses functionality. Artificial intelligence techniques, such as machine learning and fuzzy logic are used for classification of the EOG signal, but they need calibration parameters that are obtained through databases generated through prior user training, depending on the effectiveness of the algorithm, the learning curve, and the response time of the system. In this work, by optimizing the parameters of the EOG signal, the classification is customized and the domain time of the system is reduced without the need for a database and the training time of the user is minimized, significantly reducing the time of the learning curve. The results are implemented in an HMI for the generation of points in a Cartesian space () in order to control a manipulator robot that follows a desired trajectory by means of the movement of the user's eyeball.
尽管不同的生理信号，如眼电图（EOG）已被广泛应用于残疾人辅助系统的控制中，但定制信号分类系统仍然是一个挑战。在大多数接口中，用户必须适应分类参数，尽管理想情况下系统必须适应用户参数。因此，在这项工作中，提出了使用多层神经网络（MNN）将EOG信号建模为数学函数，并使用遗传算法对其进行优化，以获得每个人EOG信号的最大和最小幅度阈值，从而校准设计的接口。生理信号电压阈值变化的问题通过每3分钟执行一次的智能校准来解决；如果辅助系统未校准，它将失去功能。人工智能技术，如机器学习和模糊逻辑，用于EOG信号的分类，但它们需要通过先前用户训练生成的数据库获得校准参数，这取决于算法的有效性、学习曲线和系统的响应时间。在这项工作中，通过优化EOG信号的参数，实现了定制分类，减少了系统的训练时间，无需数据库，显著缩短了学习曲线的时间。结果在人机界面（HMI）中实现，用于在笛卡尔空间中生成点，以便通过用户眼球的运动来控制遵循期望轨迹的操纵机器人。
