The Pairs Network of Attention model for Fine-grained Classification

This paper proposes a novel pairs network of attention model for fine-grained image classification. The pairs network can randomly select paired inputs from dataset to compare the difference via the hierarchical-attention features learning. The loss function also considers the losses of difference in paired images according to the intra-variance and inter-variance. The pairs network and hierarchical attention feature are jointly employed to simultaneously remove noise patches and retain salient features. To evaluate the performance, the proposed pairs network is implemented in disaster-scenes classification. The disaster-scene dataset, which is collected via remote sensing imagery, has the complex scenes and includes multiple type of disasters. Compared with other methods, experimental results demonstrate the attention-model pairs network is robust to various datasets and has the better performance.