Overview: With the development of the computer and the artificial intelligence technology, the orbital angular momentum shift keying system decoding method based on the machine learning has emerged. The orbital angular momentum demodulation scheme using machine learning has advantages of the simple structure, wide recognition range and high recognition accuracy. The development of deep learning has further improved the recognition accuracy of orbital angular momentum. And the development of deep learning has further improved the recognition accuracy of orbital angular momentum. In order to speed up the training speed of the orbital angular momentum beam recognition model based on deep learning, this paper proposes to use the transfer learning method to identify the orbital angular momentum beam, and build the transfer learning recognition model based on the VGG16 architecture. To simulate the transmission of orbital angular momentum beams in a turbulent environment, this paper use the sub-harmonic method to generate an atmospheric turbulence phase screen and build a simulated turbulent environment by loading the phase screen with the spatial light modulator. The orbital angular momentum recognition task was carried out in a weakly turbulent environment with D/r₀=1.5 and a medium turbulent environment with D/r₀=4. And high recognition rates of 98.62% and 94.37% were obtained in weak turbulence environment with D/r₀=1.5 and a medium turbulence environment with D/r₀=4, respectively. The feasibility of an orbital angular momentum recognition system based on the transfer learning is proved. At the same time, in terms of the model training speed and recognition rate, this paper compares the performance of the transfer learning model and the original VGG16 model, and visualizes the recognition results of each beam by using the confusion matrix. The VGG16 model obtains the recognition rates of 99.39% and 94.81% in the weak turbulence environment with D/r₀=1.5 and the medium turbulence environment with D/r₀=4, respectively. The recognition rate is reduced by less than 1%, but the model training speed is improved by 2.3 times. This paper proves the feasibility of the orbital angular momentum recognition system based on transfer learning. At the same time, it is proved that the orbital angular momentum recognition system based on transfer learning model can greatly reduce the time required for model training under the condition of maintaining high recognition rate. This paper provides an idea for the rapid construction of orbital angular momentum shift keying system which based on convolutional neural network in the future. This paper proposes a transfer learning method to recognize the orbital angular momentum beam to speed up the training speed of the orbital angular momentum beam recognition model based on deep learning. In order to simulate the atmospheric turbulence, we generate the atmospheric turbulence phase screen by the sub-harmonic method and build the simulated turbulence environment by loading the phase screen on the spatial light modulator. The orbital angular momentum beam recognition system based on transfer learning has achieved a recognition rate of more than 90% in both weak and medium turbulent environments. Compared with the traditional deep learning method in the aspects of model training speed and recognition rate, it is proved that the orbital angular momentum beam recognition method based on transfer learning can reduce the training time while maintaining a high recognition rate in the weak and medium turbulent environment.