The transmission of vortex beam in wireless optical communication is susceptible to atmospheric turbulence (AT) and oceanic turbulence (OT), leading to crosstalk among different modes and posing significant challenges in the recognition of orbital angular momentum (OAM) modes. This study introduces a novel integration of transfer learning with the VGG16 neural network to form a transfer learning method, which is adeptly applied to the recognition of OAM modes between AT and OT speckle images. The model is trained the model with a substantial amount of data from AT speckle dataset, and then test the performance of recognizing OAM modes using OT speckle dataset. The recognition accuracy, precision, recall and F1-score for both 5 OAM modes and 8 OAM modes using the method are studied, with additional exploration into the impact of training with hybrid datasets on the model’s efficacy. The results show that the transfer learning method is effective to recognize OAM modes from speckle images affected by OT. It is observed that training with a mixed dataset fosters a greater degree of similarity between the source domain data and the target domain data, thereby enabling the model to capture intricate features and enhance recognition accuracy. The approach delineated in this paper highlights the versatility of employing transfer learning method for OAM modes recognition across analogous domains, offering a robust solution for effective recognition of OAM modes.