Enhancing Aedes aegypti breeding site detection: A deep learning approach to tire identification in unmanned aerial vehicle imagery from urban Indonesia

Discarded tires remain a significant public health problem in tropical urban environments, serving as high-risk breeding sites for Aedes aegypti, the mosquitoes which transmit dengue, chikungunya, and Zika viruses. Identifying these habitats using ground-based surveillance methods can be resource-intensive and often misses fenced-off areas and tires on roofs. This study presents a novel approach using unmanned aerial vehicle (UAV) imagery combined with deep learning to automatically detect discarded tires across the Tallo sub-district of Makassar, Indonesia, an area with historically high dengue incidence. We trained and compared U-Net++ and DeepLabV3++ convolutional neural network architectures for tire segmentation. We evaluated the models’ ability to identify the locations, shapes, and spatial distribution of discarded tires in the study area. Both models demonstrated high prediction accuracy, with F1 scores of 0.87 and 0.82, respectively, with modestly superior performance by the U-Net++. The strong detection performance was largely attributable to the distinctive circular morphology and spectral characteristics of tires in aerial imagery, though this same characteristic occasionally led to false positives when encountering tire-mimicking objects such as outdoor air conditioning condenser units, water tank lids, and similar circular items. Identifying, locating, and eliminating potential larval breeding sites is a primary strategy for the control of Aedes aegypti -transmitted viruses. Accurate, rapid detection of high-risk Aedes aegypti breeding sites from UAV imaging is a promising new tool to improve vector control.