Author(s)

Rui Xu, Runlin Liu

Corresponding Author

Rui Xu

ABSTRACT

With the development of AI technology, machine learning technology has been widely used in image recognition and classification. While most applications are for human features, some advances have been made for animals like dogs. Dogs are common animals in the world, and they play important roles in human societies. Thus, there is a wide range of applications for the use of AI in recognizing and classifying dog breeds. In this paper, we propose two fine-tuned pretrained models (InceptionV3 and ResNet50) to classify dog breeds using convolutional neural network (CNN) with transfer learning. We remove the fully connected layers of InceptionV3 and Resnet50, as well as add optimization layers, so that it is robust and can prevent over-reliance on the classification. The experimental result shows that the fine-tuned InceptionV3 can achieve 89.3% accuracy on the classification of 120 dog breeds, while the fine-tuned ResNet50 can achieve 80.1% accuracy.

KEYWORDS

Dog breed classification, InceptionV3, ResNet50, Deep learning, CNN, Transfer learning