Monitoring of flow regimes in aerated stirred tanks is important to ensure energy efficiency and product quality. The use of deep learning models for the recognition of flow regimes shows promising results. However, such models require a large amount of data for training. The aim of this paper is to apply the deep transfer learning approach to address this challenge. We compare various pre-trained models with the differential learning rate and 2-step transfer learning approaches to analyse the resultant model performance. We also investigate the effect of the dataset size on the classification accuracy.