Neural Network Identifies Dangerous Mosquitoes
Not all mosquitoes are created equal – but telling where the most dangerous, disease-spreading mosquitoes are proliferating is often a difficult task. This difficult task is also necessary for the containment of viruses like dengue fever, Zika, and yellow fever. Now, a team from the Open University of Catalonia (UOC) has developed a neural network to identify one of those problem species: tiger mosquitoes.
“Entomologists and experts can identify mosquitoes in the laboratory by analysing the spectral waveforms of their wing beats, the DNA of larvae and morphological parts of the body,” said Gereziher Adhane, a PhD student in the SUNAI Lab at UOC. “This type of analysis depends largely on human expertise and requires the collaboration of professionals, is typically time-consuming, and is not cost-effective because of the possible rapid propagation of invasive species. Moreover, this way of studying populations of mosquitoes is not easy to adapt to identify large groups with experiments carried out outside the laboratory or with images obtained in uncontrolled conditions.”
“Automated systems to identify mosquitoes could help entomologists to monitor the spread of disease vectors with ease,” Adhane said. So, instead, the UOC research team turned to neural networks – specifically, two deep convolutional neural networks. They trained these neural networks with data captured by citizen scientists through Mosquito Alert, a citizen science project run by the Centre for Research on Ecology and Forestry Applications, the Blanes Centre for Advanced Studies, and the Universitat Pompeu Fabra. “Mosquito Alert is a platform set up in 2014 to monitor and control disease-carrying mosquitoes,” Adhane said. Through Mosquito Alert, civilians can use their phones to take photos of mosquitoes and their breeding grounds, allowing entomologists and epidemiologists to track the spread of mosquitoes.
Using this data, the researchers got to work training the neural networks to identify tiger mosquitoes – but it wasn’t entirely smooth sailing. “The greatest challenge we encountered in identifying the type of mosquito in this study was due to images taken in uncontrolled conditions by citizens,” Adhane said, explaining that many images weren’t close enough to the subject, contained additional objects that confused the neural networks, or pictured squashed mosquitoes, which – for obvious reasons – prove somewhat more difficult to identify.
After some tinkering, the deep neural networks proved more than capable enough for the task: the tool achieved a reported testing accuracy of 94%. “The neural network we have developed can perform as well or nearly as well as a human expert and the algorithm is sufficiently powerful to process massive amounts of images,” Adhane said. Much work remains to be done, but, Adhane concluded, “it is possible to make predictions about images of mosquitoes taken using smartphones efficiently and in real time[.]”