Classification of phytoplankton digital holograms using transfer learning

Phytoplankton, as an important component of marine ecosystem, play a key role in understanding marine resources and monitoring marine environment. In this paper, under the suggestion of marine biology professional researcher, we selected six kinds of phytoplankton commonly found in China’s coastal areas and conducted digital holographic microscopic imaging experiments to obtain their holograms. Then, the angular spectrum reconstruction algorithm was used to conduct diffraction reconstruction of the phytoplankton hologram to achieve clear imaging of the algae target. A threshold based image segmentation algorithm is used to segment the phytoplankton target area and obtain image dataset. Finally, transfer learning is used to train on the pre-trained model. Experimental results show that the classification accuracy of the trained network on the test set can reach 95.7%.