Trusted Fine-Grained Image Classification through Hierarchical Evidence Fusion

Fine-Grained Image Classification (FGIC) aims to classify images into specific subordinate classes of a superclass. Due insufficient training data and confusing samples, FGIC may produce uncertain classification results that are untrusted for applications. In fact, can be viewed as hierarchical process the multilayer information facilitates reduce uncertainty improve reliability FGIC. this paper, we adopt evidence theory measure confidence in propose trusted method through fusing evidence. Comparing with traditional approaches, not only generates accurate but also reduces fine-grained classification. Specifically, construct an extractor at each layer extract (multi-grained) image To fuse extracted multi-grained from coarse fine, formulate fusion Dirichlet hyper probability distribution thereby hierarchically decompose coarse-grained enhance performances. The ablation experiments validate precision comparison state-of-the-art methods shows our proposed achieves competitive