Researchers from the Universitat Oberta de Catalunya’s (UOC) Faculty of Computer Science, Multimedia and Telecommunications‘ Scene Understanding and Artificial Intelligence (SUNAI) research group have developed a method for learning to identify mosquitoes using a large number of images taken with mobile phones and uploaded to the Mosquito Alert platform.
Investigating and controlling disease-carrying mosquitoes through citizen science
Mosquito Alert is a citizen science project set up in 2014 coordinated by the Centre for Research on Ecology and Forestry Applications, the Blanes Centre for Advanced Studies, and the Universitat Pompeu Fabra, to which UOC researchers have contributed.
This project pulls together data gathered by citizen volunteers who use their mobile phones to photograph mosquitos and their nesting locations in public spaces. The location of the observation, as well as any required details to aid in the identification of the species, are also collected along with the photo. Entomologists and other professionals then analyze the data to confirm the existence of a potentially disease-carrying species and notify the appropriate authorities. Citizens may help develop a map of mosquito distribution around the world and attack them using a simple photo and an app in this way.
The greatest challenge in identifying the type of mosquito in this study was images taken in uncontrolled conditions by citizens.
In the laboratory, entomologists and scientists may identify mosquitoes by analyzing the spectral wave patterns of their wing beats, larval DNA, and morphological components of the body. Because of the potential for invasive species to spread quickly, this form of analysis relies heavily on human skills and requires the collaboration of professionals. It is often time-consuming and not cost-effective.
Furthermore, this way of studying mosquito populations is difficult to adapt to detect big groups using non-laboratory trials or photos taken under uncontrolled situations. This is where neural networks could come into play as a practical approach for mosquito control.
Deep neural networks, cutting-edge technology for studying/detecting mosquitoes
Because the data in the Mosquito Alert platform contains numerous details and there is a high degree of similarity between the morphological features of different mosquito species, traditional machine learning algorithms are ineffective for big data analysis.
However, using images posted to the platform, the UOC researchers demonstrated that deep neural networks can be utilized to discern between the morphological similarities of various mosquito species. The neural network developed can perform as well as or nearly as well as a human expert, and the algorithm is powerful enough to process large volumes of data,” according to the researchers.
How does a deep neural network work?
When a deep neural network gets input data, it learns information patterns through convolution, pooling, and activation layers before reaching the output units, which perform the classification task.
“There has to be some form of feedback for a neural network to learn in order to lessen the disparity between real values and those anticipated by the computer process. The network is trained until the designers are satisfied with its performance. With minor tweaks, the built model might be used in practical applications, such as mobile apps.
Despite the fact that there is still more work to be done, the researcher concludes that “with this trained network, it is possible to make predictions regarding photographs of mosquitoes captured with smartphones in real-time, as has happened with the Mosquito Alert project.”
The post A Deep Convolutional Neural Network For Classification Of Tiger Mosquitoes appeared first on MarkTechPost.