Transforming mosquito surveillance LarvaFormer as a hybrid deep learning tool for rapid and accurate larval recognition

Mosquito-borne diseases represent a significant global health threat, causing over 700,000 deaths annually. Traditional control methods targeting adult mosquitoes are often ineffective and environmentally harmful. This paper proposes LarvaFormer, a novel hybrid deep learning framework that integrates convolutional neural networks (CNNs) and Vision Transformers with specialized attention mechanisms for the automated classification of mosquito larvae. Leveraging two benchmark datasets MosquitoLarvae-7400 (four species) and MLMI-2024 (anatomical-level imaging) the model achieves state-of-the-art accuracies of 98.97% and 99.50%, respectively. LarvaFormer incorporates cross-modality feature fusion, a lightweight attention module, and a neural architecture search (NAS)-guided optimization scheme to enhance discriminative capability and generalization. Comprehensive evaluations, including ablation studies and Grad-CAM visualizations, demonstrate its robustness, interpretability, and efficiency. The proposed system enables rapid, accurate larval identification, supporting targeted and environmentally sustainable mosquito control strategies.