Task-adaptive small sample image classification method based on meta transfer learning

Abstract

The invention belongs to the technical field of computers, and particularly relates to a task-adaptive small sample image classification method based on meta transfer learning. According to the method, by combining meta transfer learning, the problem that the feature extraction of the MAML model by adopting the 4Conv shallow network is insufficient is solved; trainable parameters are added to learn use of balance meta-knowledge in each task, and the problems of task imbalance, category imbalance and distribution imbalance of small sample learning in a real scene are solved. According to the method, the samples with low accuracy in each task are selected, and the data of the samples are recombined, so that the samples become more difficult tasks, and the accuracy of the model is improved in the process that the meta-learner learns the more difficult tasks. According to the difficult task mining algorithm provided by the invention, samples with poor classification effects are collected online to form the difficult tasks, so that a learner can learn more quickly in the difficult tasks, and the effect is better.

Description

Technical field [0001] The present invention belongs to the field of computer technology, specifically relates to a task-based meta-transfer learning adaptive small sample image classification method. Background [0002] Deep learning has been a huge success in a variety of recognition tasks, but deep neural networks have many parameters and require enough labeled samples to train the model, which severely limits their scalability. For many rare classes, collecting a large number of training samples is not feasible, on the contrary, people often only need to see some examples of new things to quickly identify a new object class, inspired by this learning ability of humans, researchers hope that machine learning models can also be used to learn quickly in the case of few training samples, and the key to this rapid learning ability is the acquisition of prior knowledge, that is, a large number of tasks are required to train models. Enabling the model to continuously summarize the ...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to solve the problem of insufficient feature extraction of meta-learning method using shallow network to extract features and the existing small-sample learning method does not consider the imbalance problem in the real scene, and provides a task-adaptive method based on meta-transfer learning Few Shot Image Classification Methods

Images