Deep Learning-Based Automatic Identification System for Marine Plankton

Abstract: Due to their diverse species and complex morphology, the automatic identification of marine plankton has always been a challenging task. In response to this problem, this study proposes an innovative deep learning-based intelligent identification system. The system integrates advanced computer vision and artificial intelligence technologies, possessing the capability of automatic feature extraction and pattern recognition. We extensively collected marine sampling data and combined it with public datasets to construct a large-scale database containing 200 species of plankton and 500,000 images. Through the training and optimization of convolutional neural network models, the system demonstrates outstanding classification performance of 94.2% on the test set. Not only does the system outperform traditional methods in accuracy, but its inference speed has also been optimized through model compression and parallel computing, capable of processing 620 images per second. By deploying containerization and automatic scaling strategies, the system ensures high availability. This research provides strong technical support for marine ecological conservation, fisheries management, and other related fields.