Page 28 - 《水产学报》2025年第11期

朱国平，等 水产学报, 2025, 49(11): 119102

Application of machine learning in fish species identification and
stock discrimination

ZHU Guoping 1,2,3,4* , CAO Dan 1,2,3 , CHEN Yuwen 1,2,3
1. College of Marine Living Resource Sciences and Management, Shanghai Ocean University, Shanghai 201306, China;
2. Center for Polar Research, Shanghai Ocean University, Shanghai 201306, China;
3. Polar Marine Ecosystem Laboratory, Ministry of Education Key, Laboratory of Sustainable Exploitation of
Oceanic Fisheries Resources, Shanghai Ocean University, Shanghai 201306, China;
4. National Engineering Research Center for Oceanic Fisheries, Shanghai Ocean University, Shanghai 201306, China

Abstract: Fish plays an important role in the marine ecosystem and is one of the main sources of protein for humans. One of
the key issues in fishery resource exploration is the accurate identification of fish species and the correct discrimination of fish
stocks. In the context of big data, machine learning techniques as emerging data processing techniques have gradually replaced
traditional methods. Compared with traditional data analysis, machine learning has shown the advantages of high accuracy,
high robustness and high efficiency while dealing with massive and high-dimensional ocean data. Its advantages are gradually
recognized in the field of marine biology and ecology. This review firstly introduces that the current focus of fish study which
has migrated to machine learning, then summarizes the applications of machine learning in fish species identification and stock
discrimination in terms of data sources, feature selection, and classifiers. This review then introduces application scenarios of
various deep learning neural networks, with Convolutional Neural Networks as representative, in fish species identification. The
advantages and disadvantages of each classifier and the traits of fish species that suits to those classifiers are summarized from
the perspective of predictability, expandability, and data sensitivity. Finally, common metrics for currently evaluating the
effectiveness of models are summarized. The characteristics of ecological resource data and the development status of deep
learning in the era of big data are synthesized, and the problems and challenges of the applications of machine learning in fish
species identification and fish stock discrimination are also summarized.
Key words: fish; machine learning; population; species identification; neural networks; deep learning

Corresponding author: ZHU Guoping. E-mail: gpzhu@shou.edu.cn
Funding projects: Shanghai Top-tier Talent Program of Eastern Talent Plan (BJKJ2024059); National Key Research and
Development Program of China (2023YFE0104500)

https://www.china-fishery.cn 中国水产学会主办 sponsored by China Society of Fisheries
14