Deep Learning-Based Architectures for Recognition of Cow Using Cow Nose Image Pattern

Abstract

Stacked denoising auto-encoder and deep belief network are proposed as methods of deeplearning for cow nose image texture feature extraction, and for learning the extracted features forbetter representation. While stacked denoising auto-encoder is applied for encoding and decodingof the extracted features, a deep belief network is applied for learning the extracted features andrepresenting the cow nose image in feature space. Stacked denoising auto-encoder and deep beliefnetwork help in animal biometrics. Biometrics emanated from computer vision and patternrecognition and it plays an important role in the automated animal registration and identificationprocess. Using the visual attributes of cow, and for the fact that the existing visual featureextraction and representation methods are not capable of handling cow recognition; deep beliefnetwork and stacked denoising auto-encoder are proposed. An experiment performed underdifferent conditions of identification indicated that deep belief network outshines other methodswith approximately 98.99% accuracy. 4000 cow nose images from an existing database of 400individual cows contribute to the community of research especially in the animal biometrics foridentification of individual cow.