Automatic Detection and Counting of <i>Nannochloropsis oceanica</i> Microalgae: A Comparative Study of Classical Machine Learning Versus Deep Learning–Based Approaches

ABSTRACT The Nannochloropsis oceanica microalgae contain a high content of polyunsaturated fatty acids, amino acids, and proteins, which are essential in diets intended for the cultivation of mollusks, crustaceans, and fish to ensure good larval growth and better survival. Counting any microalgae is fundamental to maintaining the good health of mollusks and other species that base their diet on this organism. Currently, in different aquaponics processes, counting is done manually, which is tedious for the technician and prone to errors. This study proposes an automatic counting system based on computer vision and machine learning. We describe the development of two applications. The first one allows the technician to manually count the microalgae, marking them manually to build a microalgae dataset. Once we generated the dataset, we trained classical machine learning models and compared them against a convolutional neural network (CNN). Finally, we integrated these models into a second desktop application in which the technician selects the region of interest within the microscopy image, and the app counts the microalgae based on the previously trained models. We report 91% accuracy in the CNN‐based model we trained, which is an adequate percentage given the noise and variable lighting conditions of the microalgae dataset.