Exploratory approach of neural networks applied to orthomosaics for detection of tires as possible larval foci

Mosquitoes causes several pandemies nowadays, like dengue, malaria and others. In Paraguay Dengue has become an endemic disease, and every year there are prevention campaigns that attempt to reduce the proliferation of the Aedes Aegypti mosquito. Abandoned tires are a potential place for their re-production. Then it is important to detect and eliminate them. This work presents the use of an UAV (DJI Mavic 2 Pro) for image collection in a city environment and convolutional neural networks (SSD Mobile Network V2) for detection of tire images. For geolocating the tires two approaches were developed, the first approach is to apply object detection with the trained network to an orthomosaic. Because of a poor performance a second approach was proposed, the use of a a super-resolution network to the orthomosaic and then the object detection with the trained network. This later approach achieved better results.