Identificação biométrica de bovinos baseada no padrão do espelho nasal através de aprendizado de máquina

Abstract

The identification and traceability of cattle are crucial for the control of safety policies within the context of food production. By using a unique, inviolable and irreplaceable physical feature, in addition to accompanying the animal throughout its life and having its effectiveness proven, biometric identification proves to be the most reliable. This study presents an approach based on techniques of image processing and machine learning for biometric cattle identification, using muzzle print images. More specifically, a cascade classifier was used to find the region of interest. After the muzzle detection, two approaches for feature extraction were tested, the bag-of-words model and Weber’s local descriptor. Lastly, the k-NN classifier was used to classify the biometric features. The detector achieved good accuracy but failed in specifics scenarios. Both approaches for classification were adequate for biometric identification, reaching high values of accuracy, sensitivity, and specificity.