DeepSquitoes: A mobile system framework for the surveillance of disease-carrying mosquitoes
Résumé
Insects that spread diseases like malaria, chikungunya and Lyme disease
are found all over the world because of climate change, economic fluctuations,
human migration, and international trade. In this study, we propose DeepSquitoes,
a mobile system framework for insect identification and fast data dissemination,
with the goal of improving the management of public health hazards.
DeepSquitoes specialises in the quick identification of mosquitoes, which are
common in tropical areas, and can be used to monitor insect population
movements in real-time. To maximise user interaction and data accuracy, the
application includes geolocation-based identification, sophisticated
preprocessing, and specialised annotation. Image preprocessing techniques like
Gaussian Blur and contour extraction are applied on mosquito wing images to
ensure data quality. Deep learning algorithms are trained on the preprocessed
images for mosquito species classification. The image recognition model
performs well, with a 93% training accuracy and a 74% validation accuracy using
MobileNetV2 from TensorFlow. Our local dataset, which included 154 images of
eight different insect species, had a commendable recognition accuracy rate of
76%.