Descritor baseado em MFCCs para Detecção do Mosquito Aedes Aegypti

Abstract

Factors such as urbanization, global warming, and increased resistance of mosquitoes to insecticides have made combating the proliferation of Aedes aegypti challenging. The manual identification of outbreaks throughout mosquito traps involves financial and time-consuming resources. In this context, solutions based on audio processing and machine learning techniques for the automatic detection of mosquitoes through the tones produced during their flight have been investigated. Solutions based on deep learning have shown promising results; however, they require a large volume of data and have a high computational cost, making their wide adoption challenging, especially when it comes to audio capture and processing devices with low computational power, such as, for example, smartphones. Therefore, this work proposes a compact descriptor capable of and effective in detecting the presence of Aedes aegypti based on Mel-Frequency Cepstral Coefficients (MFCCs). The results show that the descriptor composed of 40 MFCCs combined with the XGBoost classifier can detect the presence of Aedes aegypti with values above 95% of accuracy, Kappa and F1-score. Consequently, we believe that our findings can support the implementation of a scalable and low-cost automatic Aedes aegypti detection system.