Computer vision driven automated counting traps for remote mosquito surveillance

Mosquitoes are the deadliest animal in the world infecting over 350 million people each year witha range of diseases. Driven by climate change and insecticide resistance this burden is expected togrow. Due to a lack of effective vaccines for mosquito-borne viruses integrated control ofmosquito populations remains the primary strategy for disease mitigation. Mosquito surveillance -monitoring an area to understand mosquito species composition abundance and spatial distribution- is critical to informing decisions about what control strategies will be most effective in specificlocations and is necessary to determine if interventions are effectively decreasing mosquitopopulations. Conventional surveillance practice relies on manual distribution of mosquito traps androutine visits to collect the specimens. Due to the resource-intensive nature of vector surveillancemany county and municipal departments of health particularly those in rural communities do nothave the capacity and capability to conduct routine surveillance. Here we propose to develop thefirst computer vision driven automated counting trap for mosquito surveillance. Building on priorwork to develop a computer vision system for automated mosquito species classification we willbuild an optical system for distinguishing between mosquitoes and non-mosquito arthropods. Imagerecognition techniques will focus on object detection in context of the new optical design andhardware specifications will be considered with the aim of transitioning development into a low- cost system to remotely transmit abundance information to community public health agencies andsupport actionable biosurveillance.