With the increasing demand for medical treatment, public security, anti-terrorism and other aspects of urban life, the vital signs detection of non-contact radar is gradually getting the attention. In this paper, an algorithm for radar vital signs detection based on EMD and neural networks is presented. Due to the non-linear and non-stationary characteristics of UWB radar echo signal, this paper utilizes the space and time scales characteristics of the EMD to decompose the signal and obtain a series of IMF. By combining the BP and IGA neural networks, it optimizes the signal and obtains the heart and respiratory signals. The experimental results show that the proposed algorithm is more accurate than the direct EMD decomposition and reconstruction of the signal, which makes up for the end effect of EMD decomposition, and has broad application prospects and research value.