| Abstract: |
Important vector borne zoonotic diseases are transmitted by different mosquito species. Mosquito surveillance needs expert entomologists and is time-consuming. Trap-captured mosquitoes are transported to the laboratory for counting and identification, and there are over 3,500 species of mosquitoes in the world. In order to improve mosquito surveillance, we evaluated the accuracy of a novel optoelectronic sensor prototype that captures the shadow of the mosquito while is being sucked into a trap. This is the first time that species, sex and age classification of mosquitoes is made with the forced flight condition of a commercial ventilatorbased mosquito trap, where the natural wing-beat is distorted. Culex pipiens, Aedes albopictus and Aedes aegypti were used to test the sensor. Various algorithms on different feature combinations were trained and optimized for machine learning to recognize automatically mosquitoes' sex, age and species. Our system was capable to distinguish between species and sex in terms of fundamental frequency, showing that the fundament frequency was higher in males than females and higher in mosquitoes of Aedes than in Culex genus. The system proposed in this study is useful for genus classification with accuracy values that ranged from 93. 83% to 95. 73%. More data and training will be necessary to optimize the sensor to better classify mosquito species of the same genus since the accuracy for Aedes genus was 76. 06%. Regarding gender identification, male and female were discriminated with more than 93. 11% of accuracy after machine learning techniques. This information will be important for arbovirus surveillance programs since the females are the unique implied in arbovirus transmission. The accuracy in terms of age ranged from 69. 81% to 90. 97%, allowing to know how old the mosquito population is, providing useful data due to the importance of the age in vector capacity which it is important to estimate the risk assessment for arbovirus diseases. |
guest ::
