This site uses cookies. By continuing to browse the site you are agreeing to our use of cookies. Read our privacy policy
AI-Powered Protection of Chinese White Dolphins
In Xiamen Bay, sleek gray-pink creatures peek out of calm waters. Meet the Chinese white dolphin, a nationally protected species and one of the rarest marine mammals. They are not just a symbol of Xiamen Bay's ecology, but a key indicator of coastal ecosystem health.
Chinese white dolphin in the Xiamen Bay (Image credit: Dr. Zeng Qianhui, Third Institute of Oceanography, Ministry of Natural Resources)
However, the habitats of these creatures are threatened by human activity, including shipping, fishing, and coastal engineering projects. Classified as vulnerable on the IUCN Red List, noise pollution, habitat loss, and entanglement in fishing gear are endangering their survival in the wild. To guide effective measures that can help their population thrive, it’s crucial to accurately identify the dolphin’s movements and the threats they face.
From Manual Monitoring to Intelligent Recognition
Over the past decade, researchers have used cameras to record when the dolphins break the surface. More than 300,000 images have powered an individual identification database, with researchers able to recognize ‘old friends’ at a glance based on the dorsal fin patterning.
However, manual monitoring cannot systematically monitor these sea creatures. Dolphins migrate and search for food. And when hunting, they stay submerged for long periods, making their movements difficult to track. After collecting data, researchers need to filter useful materials from a massive number of videos and photos to collect useful research data.
Screenshot of the model training process
Just like human faces, the dorsal fins of Chinese white dolphins have unique spots and notches, based on which individual animals can be identified.
Today, AI is making this possible. The Third Institute of Oceanography of the Ministry of Natural Resources and Huawei TECH4ALL have developed a feature recognition system for Chinese white dolphins. The platform covers image preprocessing, AI-powered inference and recognition, dorsal fin image cropping, data classification, and cloud display.
Interface of the Chinese white dolphin feature recognition system
Since its launch three months ago, the system has collected 2,820 images and videos, identified 13 individual dolphins, and improved data labeling efficiency fourfold. The AI model's accuracy for individual dolphin recognition surpasses 90%, and the recognition rate for the complex behaviors of Chinese white dolphins has reached to 85%. Each Chinese white dolphin has its own file, allowing researchers to track their status. With long-term data critical to formulating protection measures, researchers can see the number, distribution, age structure, and breeding behavior of the dolphins in Xiamen Bay, as well as the threats they face.
-
90%
individual recognition accuracy
-
2,820
images and videos collected
Building a Sustainable Future Together
Chinese white dolphins in the Xiamen Bay (Image credit: Dr. Zeng Qianhui, Third Institute of Oceanography, Ministry of Natural Resources)
Chinese white dolphins in Xiamen Bay are one of eight populations in China's southeast coastal areas.
With the continued efforts of research coupled with the application of AI, the secret lives of the Xiamen Bay population are gradually being revealed, giving hope for their continued survival.
