Mosquito larvae counting system in the natural environment using deep learning on images

It is known that some species of mosquitoes, from genera Aedes, Anopheles and Culex, have evolved to feed preferably from human beings and their life cycle occurs in urban areas. Gran Asuncion is an urban agglomerate that includes´ many cities around Asuncion, capital of Paraguay, which has an´ approximated population of 2.8M inhabitants from the 7.0M in the country, being the zone with the greatest population density of Paraguay, and so for mosquitoes population. This works consists of the development of a image acquisition system for obtaining images of larvae, and the training and test of a counting system using deep learning techniques. It has been obtained 976 images that have been labeled manually with boxes for further processing and different tensorflow2 models were trained in order to count the number of insects. The main contribution of this work is to offer images for data of the larvae in the region, obtained on field conditions, what is more, they have been tested with different models in order to validate the dataset. Results show the advantages and disadvantages of the different models taken into account. The model faster_rcnn_resnet101_v1_640×640 has been chosen for a detection system that will be implemented in the next stage, because it offers better results, given that all traning processes have already been done.