Automatic colony cell counting method of phytoplankton in microscopic images with multi-task learning

Phytoplankton are key bioindicators in aquatic ecosystems, where species distribution and cell abundance serve as vital metrics for environmental assessment. Although image-based automated identification has progressed, accurately counting colonial cells remains challenging. This paper proposes a novel method for simultaneous phytoplankton recognition and cell counting. Using ResNet50 as a backbone, deep features are extracted from microscopic images of algal colonies. These features are then fed into two parallel branches for classification and counting, respectively, with model parameters updated through alternating training. Evaluated on 16 common colonial algae species from Lake Chaohu, the proposed method achieved 99.2% identification accuracy, 93.9% average counting accuracy, with mean absolute error and mean squared error of 1.290 and 2.263, respectively. This method effectively overcomes the challenges associated with colonial phytoplankton by integrating both identification and counting into a unified framework, providing a robust tool for automated aquatic algal monitoring.