Application of Unmanned Aerial Vehicle and YOLOv5 Model to Identify Water for Risk Assessment of Dengue Fever

Dengue is a viral disease that is emerging rapidly as a pandemic in tropical and subtropical countries, primarily transmitted via mosquitoes. However, mosquito surveillance and control are challenging due to dynamic interaction between humans and landscapes. The main objective of this research is to develop a workflow for using drones in large-scale data acquisition to detect water sources that are visible in aerial imagery. Two different approaches will be addressed and their effectiveness will be evaluated through a case study. The first approach is photogrammetry-based that can build high-resolution 2D maps of these areas, but the rendering of their exploitation by filters is not precise and is dependent on the image quality. The second approach is to train deep learning algorithms to detect stagnant water using the YOLOv5 Instance Segmentation model. For the moment, about two-thirds of the water areas in the validation dataset are correctly predicted, implying that a well-trained deep learning model can effectively yield a more satisfactory result.