bachelorThesis
Identificação de bovinos pelo estudo do espelho nasal
Fecha
2018-08-10Registro en:
BORDIGNON, Alexandre Opeck de Morais; TURUTA, Anderson Hideyuki; ARIELLO, Guilherme Eleuterio. Identificação de bovinos pelo estudo do espelho nasal. 2018. 80 f. Trabalho de Conclusão de Curso (Graduação em Engenharia Eletrônica) - Universidade Tecnológica Federal do Paraná, Curitiba, 2018.
Autor
Bordignon, Alexandre Opeck de Morais
Turuta, Anderson Hideyuki
Ariello, Guilherme Eleuterio
Resumen
The identification of animals inside a production chain, in a fast and reliable way, is important in many different aspects, and for this reason it has been gaining more attention of researchers over the recent years. As an example of this, it can be cited the necessary control over cattles sent to a frigorific or to milking farms. The traditional identification systems available in the current market have various issues, as information loss and fraud possibilities. This fact, allied with the demand of quality and traceability of bovines exported, creates a necessity for an alternative to this system. Starting from this scenario, this work was developed to analyse the possibility of biometrical identification of cattles through the image of their central nasal pattern, center region, located below the nostril and above the upper lip of the animal, and the image of the whole muzzle of the bovine. In view of, as fingerprint in human beings, bovines show oneness at the biometrical characteristics at these regions. Our research investigated various algorithms of digital imaging processing, characteristics extraction and classification, as well as the possibility of moving the developed software to a embedded system over NVidia Jetson TX1 board, looking for setting a precedent to developing a real time bovines identification system. Hit percentages of 100% were obtained using a i7 computer, using characteristics extraction algorithms PCA, SIFT, SURF and a Convolutional Neural Network, and the classification algorithms kNN and SVM, when they were applied to a image set provided by the University of Sao Paulo. The ˜ same tests, when applied at the image set provided by the Agronomic Institute of Parana had a ´ hit percentage of 100% only in the case when the Convolutional Neural Network was used as an extractor of characteristics. With the NVidia board, USP’s image set had a hit percentage of 99.51%, while IAPAR image set had a hit percentage of 93.65%.