MosQNet-SA: Explainable convolutional-attention network for mosquito classification with application as a RESTful API for dengue and malaria risk mapping

Mosquito-borne diseases represent a substantial risk to public health around the world and require a precise and efficient classification of mosquitoes for effective monitoring and management. Mosquitoes spread microorganisms responsible for various infectious diseases in humans and animals, including malaria, dengue, chikungunya, and encephalitis. Throughout the year, more than one million people succumb to mosquito-borne diseases, underscoring the need for mosquito classification. Deep learning architectures, trained on extensive data sets, have recently propelled numerous application sectors. Researchers have studied the automatic classification of vector mosquitoes by image analysis for decades due to its practical applications, including early diagnosis of potential mosquito-borne diseases. This study introduces MosQNet-SA, an explainable convolutional attention network developed for the precise and efficient classification of mosquito species. This innovative method achieves a classification accuracy of 99.42\% while substantially reducing parameters, up to 10 times fewer parameters, and faster inference times compared to existing models. The datasets are compiled from four sources and organized, encompassing three species of mosquitoes. Comprehensive evaluations employing explainable AI methodologies, including saliency maps, Grad-CAM, and SHAP, validated the exceptional performance, reliability, and trustworthiness of MosQNet-SA. Using a RESTful API to deploy MosQNet-SA makes it easy to integrate it into applications for real-time mosquito classification and dynamic risk mapping of diseases like dengue and malaria spread by vectors. Through this novel, efficient, and explainable deep learning approach, this research contributes to more effective vector-borne disease surveillance, monitoring, and control efforts globally.