A Lightweight Deep Learning Pipeline for Real-Time Identification of Mosquito Larvae on Mobile Devices

Abstract

Real-time detection and classification of mosquito larvae on mobile devices still face challenges in terms of accuracy and efficiency. Manual identification is limited, thereby necessitating the development of deep learning-based systems to improve both accuracy and speed in diagnosis. This study proposes a fusion deep learning model, combining YOLOv8 for object detection and MobileNetV3‑Small for mosquito larvae classification, to enhance the accuracy and efficiency of classifying mosquito larvae into three classes—Aedes, Culex, and an ‘Unknown’ class that captures non‑Aedes/Culex larvae (e.g., Anopheles, Toxorhynchites)—as well as to support environmental health monitoring. The methodology involves using YOLOv8 for object detection and MobileNetV3 for mosquito larvae classification. The dataset comprises images of Aedes and Culex larvae and curated “Unknown” examples representing other genera. The model was trained and evaluated using deep learning techniques, and subsequently deployed in a mobile application to automatically detect and classify the larvae. The results indicate that the developed system is capable of detecting and classifying mosquito larvae with high accuracy, with YOLOv8 achieving mAP@0.5 of 0.986 and mAP@0.5:0.95 of 0.777, while MobileNetV3‑Small attained a classification accuracy of 0.962. For efficiency, the model runs in real time on mobile devices with low latency. The model also demonstrates stable performance on unseen data, confirming its potential for environmental health monitoring and its role in supporting more effective vector control efforts, as well as contributing to further research in the field of entomology.