Dog Breed Classifier Using Deep Learning

Abstract

Pattern recognition(PR) is realized as a human recognition process which can be completed by computer technology. We should first enter useful information of identifying the object into the computer. For this reason, we must abstract the recognition object and establish its mathematical model to describe it and replace the recognition object for what the machine can process [1] . The description of this object is the pattern. Simply speaking, the pattern recognition is to identify the category to which the object belongs, such as the face in face recognition. Our project is based on PR which is to identify the dog’s breed. In our project, based on 10,000+ images of 120 breeds of dogs, we use 4 methods to do the identification. Each method has a different training model. The four models are ResNet18, VGG16, DenseNet161, and Alex Net. Based on our models, we also make some improvements on the optimization methods to increase our identification accuracy. After our comparisons, we find that the Dense Net model is the best, and we take it as our prime model. Our best accuracy can be up to 85.14%.