Aplicativo móvel para a identificação automática dos principais vetores da doença de chagas no Estado de Pernambuco

Abstract

Chagas disease is considered by the World Health Organization as a neglected disease and a serious public health problem, with approximately 7 million people currently infected by Trypanosoma cruzi worldwide. The objective of this work was to develop an application for the Android platform that uses a machine learning model to help identify images of triatomines that are vectors of the Chagas disease. The model used was based on EfficientNetV2, a convolutional neural network architecture pre-trained for generic image recognition, which was enhanced with images of native vectors from Pernambuco and similar non-vector insects. The developed model was evaluated using a test dataset distinct from those used in training and validation, achieving an accuracy of 91.89%. These results indicate that the adapted EfficientNetV2 model was able to generalize well to new data. The application, named TriatoDetect, was designed with an interface to facilitate the identification of triatomines through the phone's camera or images stored on the device.