Skip to content

Prototype Used Tail-feather Microstructure to Classify Ten Raptor Species

Crested serpent-eagle in flight showing the underside of its wings and tail feathers
Research · Other research

A forensic-methods study published online on 31 May 2026 built a prototype system for distinguishing ten raptor species from microscopic structures in tail feathers. The ten classes were bearded vulture (Gypaetus barbatus), golden eagle (Aquila chrysaetos), crested serpent-eagle (Spilornis cheela), besra (Accipiter virgatus), common kestrel (Falco tinnunculus), black baza (Aviceda leuphotes), Verreaux’s eagle (Aquila verreauxii), Eurasian sparrowhawk (Accipiter nisus), northern goshawk (Accipiter gentilis) and common buzzard (Buteo buteo).

Microscopic feather structures became measurable species features

The researchers used scanning electron microscopy (SEM), which images surfaces with an electron beam at far higher magnification than ordinary light microscopy. About 1,000 cropped micrograph patches were produced. YOLOv8 was trained to locate tiny feather structures such as hooklets, cilia and ventral serrations, after which the detected structures were converted into numerical features for species classification.

These microstructures were used as identification markers. The study did not experimentally test what the differences do aerodynamically or whether they reflect particular flight styles.

The reported 90% accuracy came from the constructed test dataset

The final classification stage used logistic regression. Because real scanning-electron-microscope material from protected raptors was limited, the researchers also generated about 1,000 synthetic feature samples within species-specific ranges reported in earlier studies. These artificial samples were used to provide enough data for the classifier.

The integrated system reported 90% overall classification accuracy on its test set. That result shows that the selected microstructural features could separate the ten species within the study’s constructed dataset; it is not an independent blind validation on new feathers seized in wildlife-crime investigations.

The authors identify sparse real sampling as an important limitation. Closely related species, as well as seasonal, age and subspecies variation, can produce overlapping feather structures and misclassification. They therefore argue that much broader sampling of real feathers is needed before the system can become a robust forensic identification tool.

The intended application is wildlife forensics, where loose or damaged feathers may be evidence in cases involving poaching or illegal trade. The study is best viewed as a prototype showing that quantified tail-feather microstructure can contribute to automated species identification, rather than as a field-validated replacement for established forensic identification methods.

Sources and image credit

  1. Huang, P., Li, Y., Huang, Y., Lu, R., Xia, T., Li, D., & Liu, D. (2026). YOLOv8-Logistic integration for raptor tail feather microstructure identification. In A. Grangetto, A. G. Martínez, E. A. Castro Rojas, & A. J. Tallón-Ballesteros (Eds.), Machine learning and artificial intelligence: Proceedings of MLIS 2025, Hong Kong, 24–26 November 2025 (Frontiers in Artificial Intelligence and Applications, Vol. 420, pp. 42–51). IOS Press. https://doi.org/10.3233/FAIA260009Primary open-access conference-proceedings article first published online 31 May 2026 under CC BY-NC 4.0. The study used SEM images from ten raptor species; around 1,000 micrograph patches were produced, while the logistic-regression stage also used about 1,000 synthetic feature samples generated within literature-reported species ranges. The paper reports 90% overall test-set classification accuracy and identifies limited real sampling as an important source of uncertainty.

Image: Bendale.Kaustubh / Wikimedia Commons — Crested Serpent-Eagle in flight at Dehradun, India, 4 December 2012, showing the underside of the wings and tail; representative species image, not a study specimen · CC BY-SA 3.0 · Source

About this content: This story was produced with AI assistance within an editorial workflow developed by Wildlife Vagabond. Editorial responsibility remains with Wildlife Vagabond.How AI is used

Other research

SPECIES IN THIS STORY

Species in this story

Eurasian Sparrowhawk Accipiter nisus Explore species Common Buzzard Buteo buteo Explore species Crested Serpent-Eagle Spilornis cheela Explore species Common Kestrel Falco tinnunculus Explore species

Independent research and conservation news archive

Did you find this information useful?

Wildlife Vagabond is independently built and maintained. Voluntary support helps cover source verification, hosting and continued work on research and conservation news.

Support Wildlife Vagabond

The news archive will remain freely available.