Our smartphones are pretty darn good at recognizing our voices and the contents of what we say. Could they also be good at recognizing the sounds of different mosquitoes? Absolutely, suggest researchers from the University of Oxford. With the goal of helping identify the select mosquito breeds responsible for spreading diseases like malaria, they developed a machine-learning algorithm and app that can identify the acoustic signature of different mosquito species, and categorize them accordingly.
“We have developed a set of algorithms that go some way toward detecting the presence of mosquitos by ‘listening’ to the sound of their wingbeat,” Dr. Davide Zilli, one of the researchers on the project, told Digital Trends. “On top of this, we have developed several prototype sensors, in particular, an Android app designed to run on budget smartphones. The smartphone prototype app allows us to make mosquito recordings and upload them to our servers, where we can analyze them further and use them in the ongoing training process in our algorithm development. Collecting these recordings also goes towards another goal of the project: A database of free-flying species-specific mosquito recordings that will be made openly and freely available.”
At present, the tool developed by the team can accurately identify the Anopheles species of mosquito, responsible for spreading malaria, around 72 percent of the time. Going forward, the team hopes to extend that accuracy to cover all 3,600 different mosquito species in existence — a task that requires high-quality sound recordings for them to train their system on. To help with this, the researchers are launching a citizen science project on the Zooniverse platform to help process the raw data and label it correctly.
“This is very time-consuming and we hope to use Zooniverse volunteers to help us with this classification,” Dr. Marianne Sinka, a member of the team from the department of zoology at Oxford, told us. “These labeled data are then used to directly train the algorithm to detect mosquitoes within a variety of background environments. We would also hope to use citizen science to help with first level curation of incoming recordings when the app becomes available to the wider public.”
Additional researchers on the work included Dr. Yunpeng Li, a postdoctoral researcher in machine learning, and others. A paper describing the research, “Mosquito Detection with Low-Cost Smartphones: Data Acquisition for Malaria Research,” can be read here.