Ultra-wide band (UWB) radar, as a new method of vital sign detection, can monitor human respiratory signs in real-time and non-contact mode. It is of great medical significance. To solve the problem that the distance gate of radar echo signal cannot be extracted accurately and automatically in complex scenes. A fast time domain distance gate filtering method for radar echo signals using neural network is presented. The network structure of the fully connected neural network is analyzed and designed to improve the accuracy of real-time detection of human respiratory signs. Data is collected using PulsON-440 UWB radar module. In the actual scene, the comparative analysis of experiments is completed. The results show that this method can effectively remove the non-breathing distance gate signal. It can further improve the quality of respiratory signals and robustness of detection.