LSTM-Based Infected Mosquitos Detection Using Wingbeat Sound

Abstract

Dengue fever is one of the most important mosquito-borne disease in the world. To avoid its spread, it is necessary to detect accurately and quickly the infected Aedes mosquitoes and eliminate them. Under the hypothesis of the change of behaviors of the mosquitoes when they are infected by dengue virus, we proposed a discrimination scheme of the infected mosquitoes from the healthy ones using their wingbeat signal. We constructed acoustic chamber in which a condense and omnidirectional microphone capture sthe wingbeat signal as clear as possible under the noisy environment. The proposed scheme is based on Long-Short Term Memory (LSTM) neural networks with two LSTM layers and two Fully-Connected (FC) layers. Time-frequency representation of wingbeat signal is used as input data. We identified the Spectogram, among several time-frequency representations, as the best input data for this task. Some hyperparameters of LSTM-based proposed system are adjusted after several trials. The discrimination accuracy obtained by the proposed scheme is approximately 89.35%, which is 5% better than the previously proposed method based on machine learning techniques such as K-Nearest Neighbor (KNN) and Support Vector Machine (SVM).