A Three-Stage Detection, Segmentation, and Feature Analysis Framework for Unique Individual Identification in Flamingo Populations
This study proposes a three-stage framework combining YOLOv8m for detection, SAM for segmentation and feature extraction, and PCA-enhanced Euclidean distance analysis for global ID assignment, achieving 96.4% mAP50 and enabling unique identification of individual flamingos in aerial video streams.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer
Counting wildlife in the wild is a task that has long relied on human eyes and patience, often requiring researchers to trek through difficult terrain or wait for hours to catch a glimpse of elusive animals. For species that gather in massive, shifting crowds, this traditional approach becomes nearly impossible, leading to errors where the same animal is counted multiple times or missed entirely. To solve this, scientists are increasingly turning to drones and artificial intelligence, tools that can hover above a scene and process thousands of images in seconds. However, a specific challenge remains for animals that look almost identical to one another. While some species have unique stripes or spots that make them easy to tell apart, many others, such as flamingos, present a uniform sea of pink feathers and curved necks. In a dense colony, distinguishing one bird from another without physically tagging them has been a significant hurdle for conservationists trying to monitor population health and breeding success.
A new study addresses this problem by developing a three-step system designed to identify individual flamingos within a flock without ever touching them. The researchers, working with data collected from the İzmir Delta Wetland and Bird Paradise in Turkey, utilized a drone to capture video footage of the birds. The first step of their system involves a computer program that scans the video to find every flamingo in the frame. The team tested two different versions of this detection software, comparing how well they could spot the birds at various heights. They found that flying the drone higher, at an altitude of 30 meters, actually helped the computer see the birds better. This was a surprising result, as one might expect higher altitudes to make objects look smaller and harder to identify. The reason for this improvement is that at lower heights, the birds are so tightly packed that they block each other from view, creating a confusing tangle of overlapping shapes. By flying higher, the drone captures a wider view where the birds are more spread out, allowing the software to distinguish one from another more clearly.
Once the software has located the birds, the second step begins. The system uses a specialized tool to cut the image of each flamingo out from the water and sky behind it, creating a clean silhouette. This separation is crucial because it allows the computer to focus solely on the bird's features without being distracted by reflections on the water or shadows. From these isolated images, the system extracts four types of visual information: the bird's overall shape, the texture of its feathers, its specific shade of color, and the details of its edges. Although flamingos appear uniform to the human eye, the researchers discovered that these subtle differences are enough to tell individuals apart. To determine which of these features was the most useful for identification, the team used a mathematical method to weigh their importance. They found that the shape of the bird was the most powerful clue, accounting for half of the ability to distinguish one bird from another, followed by color, edge details, and texture.
In the final stage, the system compares the unique combination of these features for every bird it sees. It assigns a permanent digital ID to each individual, ensuring that if the same flamingo appears in a different frame or at a different time, the system recognizes it as the same bird rather than counting it twice. This process effectively creates a census of unique individuals rather than just a tally of moving shapes. The study demonstrated that this method could successfully track flamingos in a dense colony covering an area of about one hectare, achieving a high level of accuracy in both finding the birds and identifying them individually. The researchers noted that their approach works particularly well on powerful computer hardware, processing video at a speed of 14.51 frames per second, which is fast enough to handle real-time monitoring.
The implications of this work extend beyond just flamingos. The researchers suggest that this framework could be adapted for other waterbirds and even terrestrial animals that lack obvious markings. By proving that unique identification is possible even for species that look nearly identical, the study offers a new path for conservationists. It moves the field away from the need for physical tags, which can stress animals, toward a non-invasive method that relies on the subtle, natural variations present in every individual. As climate change continues to alter habitats, having a reliable way to count and track specific animals in large groups will become increasingly vital for understanding how wildlife populations are faring and for planning effective protection measures.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.