Korea Disease Control and Prevention Agency Chungju, Ch'ungch'ong-bukto, Republic of Korea
Abstract: To management of mosquito-borne disease, mosquito surveillance including identification and population density is very important to decide vector control strategy. However, that is very time consuming work and necessary special knowledge. Because of that reason, we try to development mosquito surveillance system using artificial intelligence (AI) technique for proper and well-timed mosquito surveillance data. To make a mosquito image, we develop several devices with mosquito attraction, collection, freezing, and take a photograph. For the mosquito identification, we trained image classification algorithm with dominant mosquito species including Culex pipiens complex, Anopheline, Aedes albopictus, Aedes vexans, Culex tritaeniorhynchus and non-mosquito species. This architecture is built by integrating the Swin-transformer backbone with the Fast R-CNN. The classification accuracy of single mosquito individual was 97.2%. After the classification, the result was transmit to internet server with the identification result and images. This novel mosquito surveillance system has conducted a field inspection. With these monitoring system, we can identify the proper mosquito control threshold and time.
