Aiming at the problems of the noise impact on the parametric image of hand gestures, the difficulty of gesture feature extraction, and the inefficient utilization of continuous gesture time sequential information, we propose a time sequential inflated 3 dimensions (TS-I3D) convolutional neural network approach for hand gesture recognition based on frequency modulated continuous wave (FMCW) radar sensor. Specifically, the FMCW radar is used to acquire the hand gesture data, and the range and speed of the gesture in each frame signal are calculated by 2 dimensions fast Fourier transform. Then, the range-Doppler map (RDM) is generated based on the relationship between motion parameters and frequency. The interference in RDM caused by people and the external environment is filtered out and the peak of hand gesture in RDM is further enhanced by wavelet transform. Finally, TS-I3D network is designed to extract the range and speed change information of the continuous gestures. The experimental results show that the average recognition accuracy rate of the hand gestures of the proposed method is 96.17%. INDEX TERMS FMCW radar, hand gesture recognition, interference filtering, deep learning, LSTM. I. INTRODUCTION With the development of advanced signal processing approaches and deep learning [1], [2], human-computer interaction has attracted much attention. Hand gesture recognition (HGR), as one of the most important way of human-computer interaction, gets rid of the limitations of people in traditional control devices, and has been widely used in various fields, i.e., smart home like switching TV channels, turning on the air conditioner or intelligent driving like answering a phone call. The existing HGR methods mainly contain three stages, namely hand gesture data acquisition, feature extraction, and gesture classification. Hand gesture data is mainly collected using optics camera [3]-[5] or radar [6]-[9]. Optical-based gesture recognition mainly applies cameras to capture gesture images, and then applies machine-learning methods [10], [11] for feature extraction and recognition. Coelho et al. [4] use Kinect to capture RGB and depth images of hand gestures, The associate editor coordinating the review of this manuscript and approving it for publication was Guan Gui.