EchoAI: A deep-learning based model for classification of echinoderms in global oceans

Abstract

Introduction: In response to the need for automated classification in global marine biological studies, deep learning is applied to image-based classification of marine echinoderms. Methods: Images of marine echinoderms are collected and classified according to their systematic taxonomy. The images belong to 5 classes, 38 orders, 145 families, 459 genera, and 1021 species, respectively. The deep learning model, EfficientNetV2, outperforms the competing model and is chosen for developing the automated classification tool, EchoAI. Then, the EfficientNetV2-based tool, EchoAI is applied to each taxonomic level. Results: The accuracy for the test dataset was 0.980 (class), 0.876 (order), 0.738 (family), 0.612 (genus), and 0.469 (species), respectively. Online prediction service is provided. Discussion: The EchoAI model and results are facilitated for investigating the diversity, abundance and distribution of species at the global scale, and the methodological strategy can also be applied to image classification of other categories of marine organisms, which is of great significance for global marine studies. EchoAI is freely available at http://www.csbio.sjtu.edu.cn/bioinf/EchoAI/ for academic use.