Abstract: In a typical mosquito surveillance and control operation, the bottleneck of mosquito identification and processing is almost always counting collected adults and or eggs. Further, when collecting eggs in the field, lighting and magnification are not always optimal to receive results in a timely matter—much less information on species. This project, as part of a larger Aedes ecology and breeding habitat NIH grant, aimed to simplify and expedite the egg counting and species identification portion of egg collections—from the field and from a laboratory colony. Through collaboration with the University of South Carolina’s Research Cyberinfrastructure program, a machine-learning-based mosquito egg counting algorithm was constructed and applied to count Aedes eggs on germination paper. Images were taken from multiple smartphones, different colored egg papers were used, various angles and magnification were tested, and a stand for phone stabilization was even printed via a 3D printer to standardize image capture. All images were counted by a laboratory technician to compare the app accuracy to human counting. We then developed a new Convolutional Neural Network (CNN) based algorithm to detect the eggs in these images. The algorithm can also detect clusters of eggs efficiently. Our algorithm outperforms many classical and deep-learning-based methods in both accuracy and computation efficiency, performing with 92% accuracy. Future directions include distinguishing Aedes aegypti and Aedes albopictus species through egg images and expanding the phone-based application to various geographies across the United States. The program is still being finalized; however, we anticipate this application to have wide-spread impacts on those in mosquito control and surveillance who need to identify and count mosquito eggs quickly for field collection and laboratory colony maintenance.
