Abstract
Understanding mosquito behavior is key to advancing research in ecology, evolution, and disease control, yet most behavioral assays rely on human-dependent methods, such as real-time observation or manual frame-by-frame annotation, limiting throughput and reproducibility. We present YOLito, a domain-generalized AI model for automated mosquito detection and behavioral quantification. Built on the Ultralytics YOLO framework and enhanced with Slicing-Aided Hyper Inference (SAHI), YOLito accurately detects multiple mosquitoes across diverse backgrounds and imaging conditions. Trained on a globally assembled dataset of 38,547 annotated images from 35 experimental setups across six laboratories and three public datasets, YOLito achieved high performance on unseen data (precision = 0.95; recall = 0.91) and generalized across mosquito species ( Aedes, Anopheles, Culex ) and assay types, including blood-feeding, sugar-feeding, and oviposition. By automating behavioral scoring, YOLito transforms traditional assays into scalable and reproducible experimental platforms. The accompanying open-source toolkit enables high-throughput extraction of metrics such as visit frequency, duration, and distance traveled, providing a standardized and extensible framework that bridges computer vision and vector biology.
Funding Information
  • Israel Science Foundation (719/21)
  • Israel Science Foundation (2388/19)
  • Israel Ministry of Science & Technology (3-16795)
  • Israel Ministry of Science & Technology (3-17985)
  • German-Israeli Foundation for Scientific Research and Development (1833/7-1)
  • USDA (VA-160212)

This publication has 11 references indexed in Scilit: