A Deep Learning Approach for Automated Identification of Triatoma infestans Using YOLOv8

Abstract

Triatoma infestans, a primary vector of Chagas disease, poses a significant public health risk in Latin America. Rapid and accurate identification of this insect is essential for both vector surveillance programs and individual-level decision-making after potential exposure. Traditional identification methods rely on manual inspection, which is time-consuming, error-prone, and dependent on expert knowledge. This study explores the feasibility of an AI-driven detection system based on YOLOv8 to automate the identification of T. infestans from images. The model was trained on a dataset of 91 manually labeled images, with built-in data augmentation techniques dynamically generating 9,100 augmented images over 100 training epochs. The model achieved high accuracy, with a mean average precision at 50% IoU (mAP@50) of 0.9588 and a fitness score of 0.6844, demonstrating its effectiveness under controlled conditions. To assess its reliability, detection examples were analyzed in varied lighting conditions and backgrounds, as well as in scenarios where T. infestans appeared alongside visually similar insects. Results show that the model can consistently detect T. infestans while avoiding false positives for other insect species, highlighting its potential for real-world deployment. This work provides a proof of concept for the integration of AI in entomological identification tasks. Future improvements include expanding the dataset, fine-tuning model hyperparameters, and adapting the system for mobile or embedded deployment to facilitate field usability. By automating T. infestans detection, this study contributes to enhanced vector surveillance efforts and better-informed responses to potential Chagas disease exposure.