Long-tail distribution image recognition method based on hierarchical learning

Abstract

The invention discloses a long-tail distribution image recognition method based on hierarchical learning. The method comprises the following steps: 1) extracting visual features of each object in long-tail distribution image data by using a pre-training model; 2) constructing a hierarchical hyper-class tree relationship between classes according to the visual features; 3) constructing a hierarchical hyper-class tree neural network model according to the hierarchical hyper-class tree relationship; 4) training a hierarchical hyper-class tree neural network model by using a hierarchical learning-based mode; and 5) identifying the long-tail distribution image data by using the trained hierarchical hyper-class tree neural network model. According to the invention, the long-tail image recognition problem is converted into the hierarchical hyper-class learning problem from easy to difficult for the first time, and the problems of class distribution imbalance and low minority class recognitionaccuracy in long-tail distribution can be effectively relieved. Besides, the method also has excellent classification and identification performance in other data environments with different imbalance degrees, so that the method has practical application value and is worthy of popularization.

Description

technical field [0001] The invention relates to the technical field of deep learning, small sample learning and unbalanced classification in machine learning, in particular to a long-tail distribution image recognition method based on hierarchical learning. Background technique [0002] The data in real life is often extremely unbalanced. A few categories contain a large number of samples, and most categories have only a small number of samples. The distribution of samples in each category of data follows the long-tail distribution. The long-tail distribution image recognition problem is actually a combination of unbalanced classification problems and small-sample learning problems. The recognition of categories with sufficient head samples can be regarded as unbalanced classification problems, and the recognition of categories with fewer tail samples It can be regarded as a few-shot learning problem. [0003] When solving the problem of long-tail distribution image recogni...

AI Technical Summary

Problems solved by technology

However, the current research on long-tail distribution image recognition is still in the preliminary stage. All current long-tail distribution image recognition methods have not improved the recognition accuracy of tail categories very well. How to model long-tail distribution image datasets is still a problem. big challenge

Images