Automatic recognition and classification of microalgae using an inception-v3 convolution neural network model

Rapid recognition of microalgae is a key technology for real-time monitoring of algae composition and understanding the cause of algae blooms in aquatic ecological environment. Despite the development of machine learning technology, microalgal machine recognition has also been a lot of research and application, but there is not much attention to the unique microscopic characteristics of microalgal cells, which makes it difficult for the existing technologies to adapt to a large number of algae in a complex water environment. This study attempted for the first time to apply a convolutional neural network (CNN) model (Inception-v3) with higher microscopic object learning ability for microalgal classification and recognition. Through transfer learning, model training and parameter optimization, the identification accuracy of 25 species of microalgae can reach more than 90%. In addition, a new solution of introducing the morphological characteristics of the colony structure into the model was proposed to improve the identification accuracy of Microcystis of different species in the same genus, which would be of great help in classifying and monitoring the microcystin production capacity of harmful algae in water. These results will provide a new solution for the machine recognition of microalgae and suggested that more attention should be paid to the introduction of artificially identified micro-characteristics of microalgal cells into model learning.

Automated summary

This study successfully applied a convolutional neural network model to recognize and classify microalgae, and proposed a new method to improve the accuracy of identifying different species of Microcystis within the same genus, providing a new solution for machine recognition of microalgae.