Mosquitoes Species Classification Using Acoustic Features of Wing Beats

Abstract

Despite medical advancements, mosquito-based diseases are still life threating and prevalent, with dengue and malaria which contributes for deaths of more than thousands of people each year. Detecting mosquitoes’ species from their wingbeats acoustic data can be very effective but it is challenging task. A model is proposed in this paper which classifies different mosquito species based on acoustic properties of their wingbeats sound. Publicly available acoustic dataset HumBug is used to test the proposed algorithm. Due to class imbalances in the data of HumBug dataset, Receiver Operating Characteristic Area Under the Curve (ROC-AUC) score and Precision Recall (PR) curves are used to test the performance of the proposed multiple species classification algorithm. Proposed algorithm achieves ROC-AUC of 0.921 and PR AUC of 0.885 for mosquito classification using ResNet-50. Furthermore, results of the proposed algorithm are compared with the other published work in the literature.