A Lightweight Convolutional Neural Network for Personal Identification Based on Code-Modulated Visual-Evoked Potentials

Abstract

At present, personal identification based on code-modulated visual-evoked potentials is increasingly attracting people's attentions. Some convolutional neural networks (CNN) have been applied to recognize biomarkers based on code-modulated visual-evoked potentials (c-VEP) for personal identification. However, the ordinary CNNs encountered difficulties in grasping the basic characteristics of c-VEP to achieve a satisfactory performance. In this study, we proposed a lightweight convolutional neural network (LCNN) to recognize the c-VEP biomarkers in the tasks of personal identification. LCNN is composed of two parallel sub-nets, which correspond respectively to two profiles of a c-VEP sample and both include two blocks. The two blocks both contain a two-step convolutional sequence. The LCNN model is fitted by minimizing the categorical cross-entropy loss function. The goal of LCNN is to specifically handle the Electroencephalogram (EEG) data in the tasks of personal identification based on c-VEP. We recruited 20 subjects to participate in our personal identification experiments based on c-VEP. In the EEG dataset of the 20 subjects, LCNN reached the recognition accuracy of 99%. The result shows that the design of LCNN is suitable for recognizing the c-VEP biomarkers.