To improve the coding efficiency and decoding correctness of vortex optical communication. In this paper, two vortex light beams carrying different low orbit Angular momentum and radial index are used to stack to produce 16 different light intensity distribution maps, which are encoded with 4-bit binary. To address the impact of atmospheric turbulence on light intensity distribution, a Vision Transformer neural network model optimized by sparse attention algorithm is proposed, and the light intensity distribution map affected by strong turbulence is used as input for training, Thus achieving accurate identification of distorted information. The simulation experiment shows that the accuracy of this model in identifying vortex beams affected by strong turbulence can reach 95.5% and it is more accurate in resolving local details. The model excelled in recognizing accuracy despite strong turbulence, showcasing its robustness and universality across wavelengths and distances.