Human Activity Recognition Based on Deep Learning Method

With the increasing demand of security defense., anti-terrorism investigation and disaster rescue., human activity classification and recognition have become a hot research topic. When a human is illuminated by electromagnetic waves., a Doppler signal is generated from his or her moving parts. Indeed., bodily movements are what make humans micro-Doppler signatures unique., offering a chance to classify human activities. Classification needs a lot of samples for training., however., in the real application., there is a certain gap between the simulated data and the real data., and the measured data is often difficult to obtain. Due to the nonstationary characteristic for human radar echoes., the spectrograms for the human activities show different micro-Doppler signatures. Therefore., we proposed a method of human activity classification based on spectrograms using deep learning techniques., including deep convolutional generative adversarial network for expanding and enriching training set and a transfer-learned deep convolutional network (DCNN) for feature extraction and classification., which is based on a DCNN pre-trained by a large-scale RGB image data set-that is., ImageNet. Finally., the simulation results verified the effectiveness of the proposed method.