Small sample fine-grained target classification algorithm based on task adaptive feature distribution

Abstract

The invention belongs to the technical field of small sample classification, and relates to a small sample fine-grained target classification algorithm based on task adaptive feature distribution, which comprises the following steps of: constructing a task adaptive module in an embedded module, and sending all categories of features in a task into a module for learning, obtaining task-level information by traversing features of all support set images in a task; the measurement module is used for calculating the similarity between samples by combining probability statistical information of the features on the basis of DN4; compared with a small sample fine granularity classification algorithm in the prior art, the method has the advantages that the accuracy is effectively improved; moreover, the result of an ablation experiment fully proves the effectiveness of a task adaptive module and asymmetric measurement, so that all types of samples in the whole task need to be considered during training.

Images