Researchers at the Kunming Dianchi Plateau Lake Research Institute are using artificial intelligence to monitor migratory birds along Dianchi Lake, applying high-definition cameras, drones, microphones, and deep-learning models to identify and count species in real time.
The initiative—sometimes described as “bird facial recognition”—has been running since October 2022 near Haigeng Dam and has achieved roughly 90 percent accuracy in classifying and logging behavior across multiple species, according to China Daily.
After two observation seasons, the system recorded the main flock of black-headed gulls arriving about ten days later in 2024 than in previous years. Deployed at several lakeside sites, the setup has now identified seventeen species and collected hundreds of thousands of images, videos, and audio clips.
A paper published in Journal of Environmental Management in May 2025 reportedly validated the approach as part of China’s broader effort to digitize biodiversity monitoring. Similar AI-based systems have also been introduced at Chongqing’s Shuanggui Lake and the Yellow River Delta reserve, together logging more than 1,200 bird species since 2022.
Despite the nickname, the technology does not recognize individual birds. Instead, it performs species-level recognition by comparing visual and acoustic traits—such as plumage, size, beak shape, and call signatures—against trained data classes. This enables non-intrusive measurement of migration timing, population size, and behavioral patterns like feeding and roosting, generating continuous ecological data without tagging animals.
Automated year-to-year tracking of arrival dates offers early insight into climate and habitat change. Researchers say the system’s remote-sensing design also reduces physical disturbance to flocks, producing consistent records while minimizing human presence. Because the method can be replicated across sites, conservation parks and wetlands can standardize baselines and share data across provinces.
The researchers acknowledge that human review remains necessary for rare or visually similar species, especially in challenging light conditions. They note that releasing detailed performance metrics—such as confusion matrices or model documentation—would help verify generalizability and bias across habitats. For now, the Kunming deployment remains focused on species-level classification rather than re-identifying individual birds.
Source: China Daily
–
By Cass Kennedy
