AUTOMATIC CLASSIFICATION OF MOSQUITO GENERA USING TRANSFER LEARNING

Certain species of mosquitoes are the main vectors of arboviruses that cause Dengue, Yellow fever,
Chikungunya, Zika and Japanese encephalitis. These species are contained in the genera Anopheles, Culex,
Aedes. Mosquito-borne diseases pose significant threat to public health. Therefore, vector surveillance and
vector control strategies are crucial. Automation of genera identification is essential to implement effective
vector control strategies. In the past decade several machine learning and deep learning models have been
investigated for image-based automatic and accurate classification of vector mosquitoes. Such applications
also aid entomologists in insect identification task. In this study, a deep convolutional neural network
technique to classify two genera of mosquitoes: Aedes and Culex based on the morphological features is
proposed. In our work, an optimization technique i.e., transfer learning using the pretrained deep learning
models has been employed. Transfer learning saves the model training time and addresses the problem of
low performance due to insufficient amount of training data. This paper presents the architecture of three
state of the art pretrained neural networks, including VGGNet, ResNet and GoogLeNet. The models were
trained with our own dataset of images of the two genera of mosquitoes. Classification performance of the
models is evaluated in terms of classification accuracy and loss during training and validation phases of
model building.