Semi-supervised few-shot image classification method based on graph co-training

Abstract

The invention discloses a semi-supervised small sample image classification method based on graph collaborative training, which belongs to the technical field of pattern recognition. The method proposes a new label prediction method-isolated graph learning. The raw data is encoded into the graph space, reducing the dependence on features during classification, and then the samples in the graph space are projected into the label space for prediction. Secondly, a semi-supervised graph co-training method is proposed to solve the small-sample problem from the perspective of multi-modal fusion by extending the isolated graph learning to a graph co-training framework with two modal features, the rotated modality and the mirror modality. The problem of feature mismatch in learning can greatly improve the classification performance of small sample images.

Description

technical field [0001] The invention relates to the technical field of pattern recognition, in particular to a semi-supervised small sample image classification method based on graph collaborative training. Background technique [0002] In the past few years, deep learning-based visual recognition methods have reached or even surpassed human level in some cases, and an indispensable factor for success is the large amount of labeled data. But in practical situations, the burden of data collection and maintenance can be heavy. Therefore, few-shot learning for the lack of labeled samples in each class has attracted more and more attention. [0003] At present, the main small sample image classification methods are as follows: [0004] (1) Small-sample image classification method based on optimization: if the supervision information is rich, it can be learned by gradient descent and cross-validated, but the number of samples for small-sample learning is small, which is not eno...

AI Technical Summary

Problems solved by technology

[0006] In order to solve the problems existing in the image classification process of the small-sample image classification method in the prior art, the embodiment of the present invention provides a semi-supervised small-sample image classification method based on graph collaborative training, by extending isolated graph learning to graph collaborative training Framework, from the perspective of multi-modal fusion to solve the problem of feature mismatch in small-sample learning, and greatly improve the classification performance of small-sample images

Images