Automatic classification of breeds of dog using convolutional neural network

Dog is a mammal that has been a friend of man for ages, it is naturally a domestic animal with a high level of phenotype differences in behaviour and morphology. Breeding and crossbreeding activities have increased the number of dog breeds globally, thereby resulting in dogs with inter breed similarities and intra breed differences thereby creating a difficulty in their classification. The American Kennel Club (AKC) classified breeds of dog into groups based on characteristic, purpose, behaviuor and uses in order to optimize the potentials in the breeds. However, most people find it difficult to identify and classify the dog breed groups. Existing works did not consider the automatic grouping of dog breeds. Hence, there is need for automatic techniques to classify dog breeds into groups with improved accuracy. This work used the concept of Convolutional Neural Network (CNN) to develop a model that will automatically classify dog breeds into group based on the American Kennel Club standard using the Stanford’s dog dataset. The developed model achieved 92.2% accuracy, 80.0% sensitivity, 95.3% specificity and 93.4% area under curve (AUC). The model’s performance is excellent compared to existing works that used the same dataset. The experimental result was validated with two classic CNN models (ResNet-50 and SqueezeNet) using the same parameters.