ABSTRACT
The Criollo horse, one of the leading horse breeds in the south of
Brazil, has been moving the market with the demand for sports
horses. The rise of the breed in recent years was motivated by the
Brazilian Association of Criollo Horse Breeders (ABCCC), with
the mission of preserving and spreading the breed in the country
in conjunction with competitions such as the Freio de Ouro. The
demand for breed optimization follows the animal’s morphological
balance, determined by its entire bone and muscle structure
characteristics that directly impact its health and performance
as an athlete horse. Computer vision has been solving several
current problems with the deep training of convolutional neural
networks. Within this context, this work proposed to evaluate the
average precision of the Mask R-CNN model for the detection and
segmentation of Criollo horses through training this neural network
model with images collected from Google’s search engine. The best
results achieved up to 99.4% average precision for detecting horses
in the images and 89.1% accuracy for the segmentation task.