Human identification by means of optoelectronic reservoir computing

Abstract

As an improvement of the traditional recurrent neural networks (RNN), the reservoir computing (RC) only needs to train one output connection weight matrix linearly, which greatly reduces the number of machine learning network calculations. The optoelectronic RC can be realized with a delay feedback loop composed of optical and electrical devices. It has the advantages of lower power consumption and faster speed than the all-electric RC scheme. At the same time, it is easier to be controlled than the all-optical RC scheme. In this paper, we propose to employ the optoelectronic RC to process radar signals to distinguish different persons in the indoor environment. The radar signal required for the simulation is referred from the IDRad data set, which contains the echo signals of the frequency modulated continuous wave (FMCW) radar, and five persons of different ages are free to move around in the room, which is close to the real scene. First, the echo signal is processed and the micro-Doppler features are extracted, and each frame corresponds to a row vector. Then, this vector is used as the input signal of the optoelectronic RC. We numerically studied the impact of parameters such as the size of the RC and the regularization coefficient in the system. Finally, the classification accuracy of five targets reaches 87%.