High-throughput and rapid classification on harmful algal bloom species based on mega image database and artificial intelligence
Abstract
Microalgae are essential components of marine ecosystems and have significant industrial applications. However, their rapid identification, especially HAB species, poses a challenge. This study constructed a comprehensive microalgae image database and developed AI-based classification algorithms to improve identification accuracy. Using DenseNet, EfficientNet, and ViT models, we achieved high classification performance, with ViT showing the best results. The study highlighted the importance of dataset size and diversity in enhancing model performance. Additionally, the trained model was applied to analyze field-collected samples, and the results were compared with those obtained from microscopic examination and metabarcoding analyses. The application of AI technology in microalgae classification provides a reliable basis for early warning and rapid response to HABs, reducing ecological and economic losses. Our findings demonstrate the potential of AI in enhancing the speed and accuracy of microalgae identification, contributing to better management and protection of marine ecosystems.