The recognition of vortex beam orbital angular momentum (OAM) is the basis for achieving high-speed communication, but the atmospheric turbulence (AT) causes distortion of phase fronts of vortex beams, which hinders the recognition of OAM modes. In this work, we propose and investigate a joint scheme (F + GS-MSVM) of combining Gerchberg-Saxton (GS) algorithm and multi-class support vector machine (MSVM) model that can achieve high recognition accuracy (RA) of OAM modes of perfect optical vortex (POV) beam which is focused(F) by a lens at the receiver under turbulence environment. We verify the feasibility of the scheme by comparing F + GS-MSVM with other three algorithms recognition performances under various turbulence-strength levels and different beam parameters in unknown turbulent environment. The simulation results show that the RA of our F + GS-MSVM model in arbitrary turbulent environments achieves more than 98%. In unknown turbulence environment, the F + GS-MSVM model can obtain a RA of 97.1% under different hollow radius, 99% under different beam waist radius, 86% under different wavelengths and 98.2% under different transmission distances. The results demonstrate the scheme is characterized by high RA, fast speed, wide range, and high portability.