Obfs-based Tor is currently the most widely used type of anonymous communication network. The identification of this type of Tor communication can directly help block a large number of Tor anonymous communication. The existing recognition technology for this type of Tor communication extracts the characteristics of the overall data stream in the Tor communication process and combines the SVM classifier, which has low recognition efficiency and is difficult to adapt to online rapid detection. Therefore, this manuscript proposes an Obfs-based Tor communication online identification method. Different from the detection technology based on two-stage filtering that requires multiple data packets for identification, the proposed method only requires one data packet to identify this type of Tor communication during the communication process, thereby significantly improving the identification efficiency. At the same time, considering that the communication data packet has been encrypted for many times, it is difficult to manually extract the format features, and deep learning is introduced to extract the hidden features of the data packet, and the recognition model is constructed to maintain the accuracy of the recognition of this type of Tor communication. In a real Internet environment, compared to real-time identification of three Tor pluggable transports using machine learning techniques, this method increases the recognition speed of this type of Tor communication by four times, while increasing the recognition accuracy from 98 % to 99.9%. Experiments show that this method has the advantages of high recognition efficiency and high accuracy, and can quickly identify Obfs-based Tor traffic in a large amount of confused traffic.