Detection and Classification of Mosquito Larvae Based on Deep Learning Approach

This paper addresses the challenge of suboptimal biological control of the Aedes aegypti mosquito, which serves as a vector for Dengue, Chikungunya, and Zika viruses. These arthropods pose a significant threat to approximately one-third of the global population annually, capable of causing severe pain, hemorrhagic fever, and brain defects in unborn children with a single bite. The research introduces a technologically effective solution employing deep neural networks (DNNs) to conduct surveys during the immature larval stage. Our approach enables automatic identification of the biological vector in the larval stage, achieving a higher accuracy of 81.7% in region-of-interest segmentation. Moreover, it classifies larvae as Aedes positive or negative with an accuracy of 97%, significantly reducing response time from days to seconds without human intervention. The proposed solution is costeffective, minimizing the need for trained entomologists, laboratories, and expensive equipment. Utilizing microscope- based image acquisition hardware, a computer with CPU hardware, and a petri dish, sample capture and analysis become straightforward. The advantages of this proposal are particularly valuable in underdeveloped countries and remote regions, where economic constraints may limit access to preventive health services and biological vector control.