Optimized Transfer Learning for Dog Breed Classification

Animal breed classification using deep learning algorithms is required in presentation arenas. In this paper, a dataset of 70 dog breeds was considered for training and testing of transfer deep learning algorithms. The used dataset is a statistically stable dataset including approximately 100 images under each category of dog breeds. Then collected dataset was trained and tested using different deep learning algorithms like Convolutional Neural Network, VGG16, ResNet, DenseNet, InceptionNet, InceptionResNet, etc, were implemented. The outcome results were compared during algorithm training and testing based on parameters like accuracy, precision, recall, and area under curve. Further, one of the best algorithms was optimized by tuning through optimization algorithms or learning rate configurations. In the future, the proposed modules will be added along with implementations in events to fulfill the requirement of real-time dog breed recognition.