Image classification method based on unsupervised domain adversarial domain adaptation

Abstract

The invention provides an image classification method based on unsupervised domain adversarial domain adaptation. According to the method, a generative adversarial network technology is applied to image classification based on unsupervised domain adaptation; in order to eliminate the misclassification rate of edge samples, a source domain sample and a target domain sample are jointly processed based on a deep dual-channel network. The method comprises the following steps: firstly, parameters of a dual-channel network are trained by utilizing a labeled source domain sample; then, domain differences are eliminated by learning and invariant features through adversarial games and a correlation alignment method; in addition, the method enables domain invariant features to have better discrimination by combining a center discrimination method and a feature similarity enhancement method, is better in classification effect, adapts to an image classification method through an unsupervised and adversarial field, and effectively alleviates the misclassification rate of edge samples.

Description

technical field [0001] The invention relates to an image processing method, specifically an image classification method based on unsupervised domain confrontation domain adaptation, and belongs to the technical field of image processing. Background technique [0002] Image classification is an image processing method that distinguishes different types of objects according to the different characteristics reflected in the image information. It uses computer to carry out quantitative analysis on images, and classifies each pixel or area in the image or image into one of several categories to replace human visual interpretation. Image classification is an important research direction in the field of data mining and a very important means of image information organization and management. Reasonable classification of paths can not only establish a corresponding information resource library for images according to category information, better provide image retrieval and managemen...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to provide an image classification method based on unsupervised domain confrontation domain adaptation, which solves the problem of domain differences and target domain class edges that are easily misclassified, and learns domain invariant features through confrontational game strategies and correlation alignment methods. Solve the domain difference; through the joint center discriminant item and feature similarity enhancement item, the learned domain-invariant features have better discriminative properties; this method makes the learned features have better intra-class compactness and inter-class reliability Separation

Images