Deep learning classification and motion-based analysis of Aedes aegypti Red-eye GSS larvae for SIT applications

Aedes aegypti is a major vector of arboviral diseases, including dengue, Zika, and chikungunya, posing significant global public health challenges. Effective control strategies, such as the sterile insect technique (SIT), require an accurate and reliable sex sorting system to ensure male only releases. Here, we present a compact deep learning framework for image-based classification of sex-linked eye-color phenotypes (black-eye males and red-eye females) and instar stages of Aedes aegypti larvae, complemented by an independent optical flow analysis of larval motor activity. Video recordings of individual larvae were used to generate annotated datasets for training a convolutional neural network (CNN) to classify four experimental groups. In parallel, dense optical flow analysis quantified larval activity, producing interpretable motion features. The classifier reliably distinguished between all groups across validation and test sets, with inference suitable for real-time deployment on low-power hardware. Motion analysis revealed systematic differences: L4 larvae exhibited higher activity than L3, and red-eye individuals showed slightly elevated movement relative to black-eye counterparts, indicating that developmental stage is the primary determinant of motility, with pigmentation exerting a secondary influence. The proposed framework provides accurate image-based classification while independently characterizing larval motor activity, thereby improving the biological interpretation of the analyzed groups. The image-based classifier offers a practical tool for automated larval sorting in SIT programs, whereas the independent motion analysis module provides quantitative features that may support future studies of mosquito biology and vector-control applications.