Automated counting and classifying Daphnia magna using machine vision

Abstract

Daphnia magna (D. magna) is a model organism widely used in aquatic ecotoxicology research due to its sensitivity to environmental changes. The survival and reproduction rates of D. magna are easily affected by toxic environments. However, their small size, fragility, and transparency, especially in neonate stages, make them challenging to count accurately. Traditionally, counting adult and neonate D. magna relies on manual separation and visual observation, which is not only tedious but also prone to inaccuracies. Previous attempts to aid counting with optical sensors have faced issues such as inducing stress damage due to vertical movement and an inability to distinguish between adults and neonates. With the advancement of deep learning technologies, our study employs a simple light source culture device and utilizes the Mask2Former model to analyze D. magna against the background. Additionally, the U-Net model is used for comparative analysis. We also applied OpenCV technology for automatic counting of adult and neonate D. magna. The model's results were compared against manual counting performed by experienced technicians. Our approach achieves an average relative accuracy of 99.72 % for adult D. magna and 98.30 % for neonate. This method not only enhances counting accuracy but also provides a fast and reliable technique for studying the survival and reproduction rates of D. magna as a model organism.