Marine plankton species classification and microbial activity analysis using deep learning

Phytoplankton blooms in the ocean degrade water quality and can lead to mass mortality of marine organisms. Therefore, accurate classification and spatiotemporal analysis of phytoplankton dynamics are crucial for protecting marine ecosystems. This study presents deep learning-based methods capable of detecting and tracking eight phytoplankton species in real-time. We employed Faster Region-Based Convolutional Neural Network (Faster R-CNN) and You Only Look Once version 4 (YOLOv4) for the simultaneous detection of multiple phytoplankton species. Additionally, the Deep Simple Online and Realtime Tracking (DeepSORT) algorithm was integrated to track the movement of phytoplankton and analyze microbial activity. Faster R-CNN achieved an average precision (AP) exceeding 0.93 for most species, while YOLOv4 attained an AP greater than 0.84. The tracking outcomes using DeepSORT revealed significant insights into the movement patterns and swimming speed variations among the phytoplankton species. These findings demonstrate the effectiveness of deep learning approaches for enhanced phytoplankton monitoring and contribute valuable knowledge toward understanding marine ecosystem dynamics.