Abstract
The rising incidence of mosquito-borne diseases represents a major global public health concern. In response, the use of Unmanned Aerial Vehicles (UAVs) combined with machine learning and deep learning techniques has emerged as a promising solution for vector surveillance and control. This work presents the development of an automatic detection model for potential mosquito breeding sites using high-resolution aerial images captured by drones. The model identifies and delineates potential mosquito breeding sites such as puddles, plant pots, buckets, water tanks, and tires. The National Institute of Public Health (INSP) of Mexico collected hundreds of high-resolution RGB images via UAV flights over the municipality of El Vergel, Tapachula, Chiapas. Furthermore, we considered a second publicly available dataset, Mosquito Breeding Grounds V2 (MBG-V2), which contains images from different regions of Rio de Janeiro, Brazil. The images were manually labeled for the training and comparative evaluation of several object detection models, including YOLOv4, YOLOv7, YOLOv8, and YOLOv12. The best-performing model was selected using standard evaluation metrics. While the approach identifies potential mosquito breeding sites rather than directly confirming water presence or immature mosquito stages, its generalization across different geographical regions, environmental conditions, and image acquisition settings remains an opportunity for future investigation. This research contributes to the development of automated tools for detecting potential mosquito breeding sites in southern Mexico, supporting vector surveillance and potentially reducing the time and resources required for disease prevention campaigns.