Acquisition of Digital Images and Identification of Aedes Aegypti Mosquito Eggs using Classification and Deep Learning
Pedro Saint Clair Garcia, Rafael M. Martins, George Luiz Lins Machado-Coelho, Guillermo Cámara-Chávez · 2019
The mosquito Aedes aegypti can transmit some diseases, which makes the study of the proliferation of this vector a necessary task. With the use of traps made in the laboratory, called ovitraps, it is possible to map egg deposition in a community. Through a camera, coupled with a magnifying glass, are acquired images containing the elements (eggs) to be counted. First, the goal is to find pixels with a similar color to mosquito eggs; for that, we take advantage of the slice color method. From these already worked images, a process of transfer learning with a convolutional neural network (CNN) is carried out. The intention is to separate which elements are eggs from the others. In 10% of the test images, the count performed by the model, and the ground truth of the number of eggs was considered weakly correlated. This problem occurs in images that have a high density of eggs or appear black elements that resemble mosquito eggs, but they are not. For the remaining 90% of the test images, the counting was considered to be perfectly correlated.