Fine-grained image classification based on the combination of artificial features and deep convolutional activation features

Abstract

Fine-grained image classification is a challenging research topic in the field of computer vision, whose goal is to identify subclasses, such as distinguishing between different kinds of dogs. In order to overcome this problem, we combine the advantages of artificial feature with deep convolutional activation feature and design support vector machines (SVM) based on the importance of features. In this paper, we first use the bilinear neural network model to extract the deep convolutional activation feature of samples and combine it with artificial feature. The bilinear form simplifies gradient computation and allows end-to-end training. Then, multi-kernel SVM based on weighted features are trained to complete the image classification task. Finally, we present experiments and visualizations on FGVC-Aircraft and Stanford Dogs databases that analyze the effects of combined features and the multi-kernel SVM on the fine-grained object classification. The 83.8% and 66.1% accuracy proves the effectiveness of our strategy.