AUTHOR=Zhou Peng , Zhou Zhinuo , Yang Chenghan , Fang Yi , Bu Yu-Xin , Wang Chun-Sheng , Zhang Dong-Sheng , Shen Hong-Bin , Pan Xiaoyong 

TITLE=Development of a multi-modal deep-learning-based web application for image classification of marine decapods in conservation practice

JOURNAL=Frontiers in Marine Science

VOLUME=Volume 12 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/marine-science/articles/10.3389/fmars.2025.1496831

DOI=10.3389/fmars.2025.1496831

ISSN=2296-7745

ABSTRACT=Marine invasive Decapoda species have caused huge losses to biodiversity and world fisheries. Early awareness of non-indigenous species (NIS) is critical to prompt response and mitigate impacts. Citizen support has emerged as a valuable tool for the early detection of NIS worldwide. However, the great biodiversity of Decapoda species in global oceans poses challenges for the public to the recognize marine Decapoda species, especially for the uncommon or unfamiliar specimens, which sometimes might be NIS. However, despite the remarkable performance of deep learning (DL) techniques in automated image analysis, there remains a scarcity of professional tools tailored specifically for the image classification of diverse decapods. To tackle this challenge, a web application for automated image classification of marine Decapoda species, termed DecapodAI, was developed by training a fine-tuned Contrastive Language–Image Pretraining model with the images from the World Register of Marine Species. For the test dataset, DecapodAI achieved average accuracies of 0.717 for family, 0.719 for genus, and 0.773 for species. Online service is provided at http://www.csbio.sjtu.edu.cn/bioinf/DecapodAI/. It is expected to promote public participation by alleviating the burden of manually analyzing images and has promising application prospects in exploring and monitoring the biodiversity of decapods in global oceans, including early awareness of NIS.