Identification of Deadliest Mosquitoes Using Wing Beats Sound Classification on Tiny Embedded System Using Machine Learning and Edge Impulse Platform
Kirankumar Trivedi, Harsh Vardhan Jonathan Shroff
Abstract
Mosquitoes are the deadliest animal on the planet, infecting about 700 million people each year and causing over one million deaths, accounting for 17% of all infectious illnesses worldwide. We are still fighting the three most deadly mosquito species, Anopheles, Aedes, and Culex, 124 years after Sir Ronald Ross made the first pivotal discovery. Mosquitoes are difficult to detect manually since they are small and fly rapidly. The auditory categorization of mosquito wing beats may be used to detect them using machine learning. This article discusses an Arduino Nano BLE 33 Sense-based prototype that collects audio data from mosquito wing beats and utilizes TinyML to automatically classify mosquito species. With 88.3% accuracy, the TinyML system developed by Edge Impulse based on the HumBug project mosquito wing beats dataset recognizes mosquito types. To conduct this research, the frequency of mosquito wing beats was graphically represented as a feature using a spectrogram. Furthermore, live mosquito detection studies using the low-cost Arduino Nano BLE 33 Sense yielded excellent results. During testing, the model had an accuracy of 88.3% and a loss of 0.26. The use of machine learning to solve the challenge of manual mosquito type identification is efficient and has the potential to have a large impact on vector-borne illness management. The model may still be fine-tuned to get more accurate results with reduced latency. In addition, the deployment went as expected.