https://doi.org/10.1177/15579018261460206

Hemmati, S., Zastepa, A., Yoon, H., & Kim, Y. (2026). Deep learning-based classification of picocyanobacteria and Microcystis using multiwavelength laser microscopy. Environmental Engineering Science, 0(0). https://doi.org/10.1177/15579018261460206

Microcystis, Synechococcus, and Cyanobium are cyanotoxin-releasing microbes in harmful algal blooms. They often coexist and share similar morphology and flocculation patterns, making it difficult to differentiate in microscopic images. Multiwavelength laser microscopy was used (405-, 488-, 561-, and 640-nm lasers with black-and-white), and a deep learning model was developed to classify cyanobacteria. The accuracy was 76.56% using only black-and-white images for machine training; however, it improved to 98.44% with all five channel images, emphasizing the importance of wavelength laser microscopy. Microscopic images of individual wavelength lasers (with black-and-white images) improved classification performance, but the improvements were insufficient for strain-level classification. For optimal model performance, 45 image sets per sample, 200 epochs, and 256 × 256 pixel images are recommended, requiring only 11 min for machine training. In conclusion, multiwavelength laser microscopy allowed very efficient training of the deep learning model as a promising step toward reliable classification of cyanobacteria in environmental monitoring applications.

https://doi.org/10.1177/15579018261460206