Abstract: The live detection algorithm based on neural network can prevent all kinds of spoofing attacks, but can not take into account the detection speed, detection accuracy and cost. Aiming this problem, we propose a multimodal data fusion live detection algorithm based on MobileNetv2. It introduced cross stage partial network (CSPNet) and DDConvBlock to achieve the design goal of balancing detection speed, detection accuracy and cost. In order to verify its effectiveness, experiments were carried out on public dataset Msspoof and CASIA-SURF. According to actual needs, a private dataset was constructed, and the same detection algorithm was used for modeling and experimental verification on this dataset. The verification results show that it can perform liveness detection under different occlusions and different spoofing attacks. At the same time, the detection model has the characteristics of portability, and can be extended to platforms with high requirements for security and convenience.