Page 118 - 《水产学报》2025年第7期

黄真理，等 水产学报, 2025, 49(7): 079309

Artificial intelligence fish-monitoring links Gymnocypris przewalskii
migration to river environment

4
1
2
1*
3
1
HUANG Zhenli , WEN Hao , QI Hongfang , YE Guanzhong , LI Haiying , WANG Luhai ,
3
2
FU Shengyun , ZHONG Nanchang , LIU Jun 2
1. China Institute of Water Resources and Hydropower Research, Beijing 100038, China;
2. Cloudwalk Technology Group Co., Ltd., Guangzhou 511458, China;
3. Key Laboratory of Breeding and Protection of Gymnocypris przewalskii,
Qinghai Naked Carp Rescue Center, Xining 810016, China;
4. Beijing Sonar-Light Vision Technology Co., Ltd., Beijing 101407, China
Abstract: The naked carp (Gymnocypris przewalskii) is a keystone and anadromous species in Qinghai Lake, but its annual
reproductive migration to rivers is not well understood due to insufficient monitoring tools. Here, we introduced a ″fish
optical video + artificial intelligence″ technology, developed an artificial intelligence fish detection (AI-Fish) algorithm
and intelligent monitoring equipment for shallow rivers. We also established a standard growth model for G. przewalskii to
estimate the age of fish and applied it to a 30-metre-wide section of the lower Quanji River. From 2021 to 2024, monitoring
data revealed the migration dynamics and environmental interactions of G. przewalskii in Qinghai Lake. The annual migration
comprises three phases: early (upstream migration), middle (breeding), and late (downstream migration). River flow and water
temperature primarily influence migration timing, with a critical flow rate of 2.5 m³/s promoting upstream migration in the
Quanji River, peaking at 10 m³/s. Upstream migration aligns with daily water temperature rhythms, peaking in the afternoon
(15:00-16:00) and evening (20:00-21:00), with a daily maximum of 362,000 individuals. Notably, not all migrants spawn,
exhibiting 'non-reproductive accompanying migration.' Migration is geographically limited by natural barriers and altitude. Big
data analysis of over 700 000 individuals showed a normally distributed age structure, dominated by 3-7 years old, with dis-
tinct male and female distributions influenced by sex ratio. Our results indicate that the developed equipment effectively can be
used for all-weather and real-time monitoring of the migratory population of G. przewalskii in Qinghai Lake, with an identifica-
tion accuracy of more than 90%, and can obtain biological parameters such as the number of fish, direction of migration, full
length, and age structure. By establishing a monitoring network in the main rivers of Qinghai Lake, we can quickly and accur-
ately collect comprehensive 'big data' on the migrating population each year, providing robust scientific support for their future
conservation.
Key words: Gymnocypris przewalskii; fish migration; artificial intelligence; river section monitoring; environmental impact
Corresponding author: HUANG Zhenli. E-mail: zhlhuang@263.net

Funding projects: National Natural Science Foundation of China (52079148); Qinghai Provincial Science and Technology
Program (2024-SF-152); China Institute of Water Resources and Hydropower Research Fund (SS0145B022021)

https://www.china-fishery.cn 中国水产学会主办 sponsored by China Society of Fisheries
16