Buzzing through Noise: A Novel Deep Learning Model for Mosquito Species Classification usingWing Beat Audio

Abstract

Despite global health advancements, mosquito borne diseases are still life threating and prevalent, with dengue and malaria contributing to thousands of deaths each year. However, the responsibility for transmitting these specific diseases does not lie with all mosquito species globally. Detecting mosquito species from their wingbeats acoustic data can be very effective, however, it is a challenging task to accurately distinguish specific mosquito species based on their wingbeat patterns. Our acoustic-based deep learning model for mosquito species classification uses the combination of sinc layer, one dimensional convolutional layer and long short-term memory (LSTM) layer to learn features from raw audio signals and capture the temporal dependencies between them strikingly outperforming existing methods. We achieve test accuracy of 93.59% precision of 0.9363, Recall of 0.9311 and F1 Score of 0.9335. Our model has significant potential for low-cost, non-invasive, and high-throughput identification of mosquito species, which can play a role on the prevention and control of mosquitoborne diseases. By utilizing the unique wingbeat patterns of mosquitoes, our model enables early identification and monitoring of disease-carrying species. This early detection allows for timely implementation of targeted interventions, focusing resources on the specific mosquito species responsible for disease transmission in a particular region. Additionally, acoustic-based models enhance surveillance systems by providing real-time tracking and monitoring of mosquito populations. This data-driven approach facilitates rapid response to potential disease outbreaks, enabling public health authorities to implement effective control measures and reduce the burden of mosquito-borne diseases on communities.