Fine-grained image classification method based on generating adversarial networks with SIFT texture input

Abstract

A fine-grained image classification method based on generating adversarial networks with SIFT (Scale Invariant Feature Transform) texture input is proposed to improve the recognition ratio of fine-grained image classification by deep learning. For the phenomenon of data sets that require a large amount of labeled information for strong supervised learning, active learning capabilities of generative and adversarial networks and excellent image modeling capabilities for target classification images are used to achieve active learning of image features. Then the difficulty of data set construction and the computational complexity are reduced, and the disturbance to the network that may be caused by manually set labeled boxes is lessened. The input method of generating the adversarial network to is fixed to balance the authenticity and diversity of the generated samples. The idea of image restoration is considered. The random input method of the generative adversarial network that combines image feature points and random noise to is used to reduce the training difficulty of the generative and adversarial network. Experiments results show that our method outperformances the current deep learning methods in fine-grained image classification.