Student Research Reports

View Research Report

As the prominence of citizen science grows and an increasing number of people are able to quickly and efficiently collect mosquito-related data, a method of classification that is accurate, fast, and simple to use becomes a growing necessity in order to be able to mitigate the spread of mosquito-borne diseases like malaria. This data is made publicly accessible through citizen science applications like the GLOBE Observer app. By using the mosquito larvae imagery stored on the GLOBE platform combined with images retrieved from public sources through the use of search engines, our study seeks to train and implement a binary image classifier that can determine, given an image of a mosquito larva, whether or not the photographed larva belongs to the genus Anopheles. Approximately 20 percent of the available GLOBE images were deemed to possess sufficient quality to form part of the classifier’s training set and therefore, our model lacks the needed accuracy and training to be available for public implementation. We hope to continue this study by building a larger training set that will not only create a more accurate model, but also contribute to other larvae identification efforts by providing a usable dataset for training any future larvae image identifiers and classifiers, and by refining the parameters and architecture of our classifier to improve accuracy. Keywords- citizen science, deep learning, image classification, Anopheles, convolutional neural network