AI learns how to recognise the species of splatted mosquitoes

Researchers have gathered 1500 images of mosquitoes – both squished and not – for training AI. Their goal is to build a smartphone app that can track the insect

By Jeremy Hsu

Artificial intelligence trained to recognise both living and dead mosquitoes could help track the three species most responsible for transmitting mosquito-borne diseases.

Mosquitoes kill more people than any other animal because they can transmit diseases such as dengue, malaria and Zika virus fever. Using AI to automatically identify different mosquito species could make it easier to track their presence worldwide – but such an AI needs many images of mosquitoes to learn from.

Song-Quan Ong at the Institute for Tropical Biology and Conservation in Malaysia and his colleague recruited three volunteers to help them image yellow fever mosquitoes (Aedes aegypti), Asian tiger mosquitoes (Aedes albopictus) and southern house mosquitoes (Culex quinquefasciatus). The researchers took two photos of each mosquito that landed on the volunteers’ hands: one right after it landed and another after it was splatted.

Some mosquitoes bit the volunteers before getting smashed but others were killed before they got the chance. “We aim to create images that are similar to real life,” says Ong.

In total, the researchers took 1500 images, half of alive mosquitoes and half of those that had been splatted.

The team then used this dataset to train two different AIs to recognise mosquitoes on human skin. The better-performing AI could guess the correct species around 80 per cent of the time. Eventually, such an AI could end up powering a smartphone app that people can use to identify mosquitoes they find and to help researchers track the insect.

There have also been other efforts to automatically identify mosquitoes. One research group previously trained an AI to classify mosquitoes according to sex, genus, species and strain among 15 species from all over the world. Another team trained an AI to identify 67 different species of mosquito. Each one built their own dataset that could, in theory, be combined with the latest one.

“The more images available of mosquitoes the better,” says Jannelle Couret at the University of Rhode Island.

Journal reference: Scientific Data, DOI: 10.1038/s41597-022-01541-w