Dengue-Infected Mosquito Detection with Uncertainty Evaluation based on Monte Carlo Dropout

Abstract

Considering Aedes mosquitoes are a principal vector of the Dengue virus causing, in the worst case, the death of infected people, accurate detection of infected Aedes mosquitoes is very important to prevent the further spread of the virus. In this paper, we propose a detection algorithm for infected Aedes aegypti mosquitoes by Dengue Virus-2 using their wingbeat signals. The proposed algorithm uses Long Short-Term Memory (LSTM) as a classifier of the input wingbeat signal into healthy mosquitoes and infected mosquitoes. All living beings, even if they are of the same species, have different characteristics depending on the season in which they are born, temperature, humidity, food, etc. This individual difference perhaps influences the level of infection, although it is fed by the same infected blood. Considering these differences between individuals, we introduce an uncertainty measure based on Monte-Carlo dropout. The proposed algorithm detects approximately 5% of uncertainty data from all input wingbeat signals in the test set and provides a classification accuracy of 94.87% without any uncertainty.