Small sample image classification method based on component supervision network

Abstract

The invention relates to the technical field of mode recognition, in particular to a small sample image classification method based on a component supervision network, and aims to solve the problem that feature extraction is not adaptive in the field of small sample image classification at present. According to the small sample image classification method based on the component supervision network, a new scheme, namely the component supervision network, for improving small sample image classification performance is provided, component information of samples is collected and multiple labels are generated by referring to a hierarchical dictionary WordNet commonly used in natural language processing, and an auxiliary task based on component supervision is constructed; a standard classification task, a component supervision auxiliary task and a self-supervision auxiliary task are used for assisting in training a feature extractor, the trained feature extractor is used for extracting features of new-class data, and a linear classifier is used for completing the final classification task so as to improve the adaptability of the feature extractor.

Description

technical field [0001] The invention relates to the technical field of pattern recognition, in particular to a small-sample image classification method based on a component supervision network. Background technique [0002] The small-sample image classification process usually includes two stages. The first stage is to use the base class data to train the feature extraction model. The second stage is to use the trained feature extraction model to extract the features of the new class data, and use the designed classifier to complete the classification. Task. Among them, the design of feature extraction model and classifier, as an important part of the small-sample image classification system, has always been the core issue in the field of small-sample image classification. [0003] At present, the main few-shot image classifier design methods are as follows: (1) Data-based few-shot image classification: Data-based few-shot learning utilizes data augmentation to enhance the ...

AI Technical Summary

Problems solved by technology

[0005] The present invention provides a small-sample image classification method based on component supervision network, aiming to solve the inadaptable problem of feature extraction existing in the field of small-sample image classification

Images