MosQuit - Acoustic Mosquito Detection and Identification

In Africa, over four million malaria cases are recorded each year, with rainfall, vector species, biting speed, and altitude all influencing transmission patterns. More than ever the world is realizing the urgency of monitoring viruses in the environment before they cause outbreaks. MosQuit will change malaria pathogens identification and also establish reliable and efficient tracing patterns.

It is a low cost TinyML powered device that attracts and identifies mosquito species. The device collects mosquitoes’ flying tones and identifies the species. As of now, it can identify the species of Anopheles, Aedes and Culex with an accuracy of 84.1%. The device captures the time, species and weather conditions for each encounter; parameters that we can utilize to create new models. When an anopheles is detected three times, an onboard mosquito spray is actuated by a servo motor.

For training the model, around 20 minutes of different mosquito sounds was used and each split into 4 seconds for training. The device can easily distinguish an anopheles mosquito with an 80% accuracy. Aedes and Culex species achieved validation accuracies of 70% after testing. In the model, Audio(MFE) was used as the processing block and Neural Network (Keras)as the learning block.

Check out the IoT Desktop Dashboard here.