When facing the challenge of multi-face recognition, the existing face recognition algorithms have the defects of low accuracy and too slow speed. This paper proposes a multi-face recognition algorithm based on the InsightFace algorithm, which is applied in the scene of large human flow throughput. First, an improved multi-task cascaded convolutional network (Multi-face-MTCNN) algorithm is proposed to accurately realize face detection and feature alignment in a multi-face environment. Moreover, we combine the MobilNetv3 structure and the GhostNet structure to propose a lightweight network structure MobileNetv3-GhostNet that is faster and can extract more facial features. The amount of parameters is greatly reduced, whose size is only 6.7Mb. This algorithm takes into account the accuracy and speed of the multi-face recognition, has good performance, and proposes new ideas for multi-face recognition.