Pet identification is important for veterinary care, pet ownership,
and animal welfare and control. This proposal presents a solution
for identifying dog breeds using dog images. The proposed method
applies a deep learning approach to identify the dog breeds. The
starting point for this method is transfer learning by retraining
existing pre-trained convolutional neural networks on the Stanford
Dog database. Three classification architectures will be used. These
classifiers will take images as input and generate feature matrices
based on their architecture. The stages these classifiers will undergo
to create feature vectors are 1) Convolution to generate feature
maps and 2) Max Pooling: highlight features are extracted from
the feature maps. Data augmentation is applied to the database to
improve classification performance.