Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Cross-domain image classification method based on coupling knowledge migration

A classification method and image technology, applied in the field of cross-domain image classification based on coupled knowledge transfer, can solve the problems of long calculation time and insufficient classification accuracy, and achieve the effect of accurate classification and reduction of differences.

Active Publication Date: 2019-10-25
GUANGDONG UNIV OF TECH
View PDF3 Cites 7 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] In order to solve the problems of long calculation time and insufficient classification accuracy of existing cross-domain image classification methods, the present invention provides a cross-domain image based on coupling knowledge transfer Classification

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Cross-domain image classification method based on coupling knowledge migration
  • Cross-domain image classification method based on coupling knowledge migration
  • Cross-domain image classification method based on coupling knowledge migration

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0063] A cross-domain image classification method based on coupled knowledge transfer, such as figure 1 shown, including the following steps:

[0064] S1. Obtain the source domain image and the target domain image and perform feature extraction, respectively obtain the source domain feature matrix as the source domain data, and obtain the target domain feature matrix as the target domain data:

[0065] S1.1. Input n s labeled source domain images and n t unlabeled target domain images; where n s , n t is a positive integer;

[0066] S1.2. Feature extraction is performed on the source domain image and the target domain image respectively, and the feature vectors of the source domain image and the target domain image are respectively extracted by any of the pre-trained VGG16, GoogLeNe, and ResNet50 neural networks, and the The feature vectors obtained by extracting the source domain image are arranged in columns to obtain the source domain data feature matrix where m is t...

Embodiment 2

[0096] In Example 2, 980 face images composed of 10 people's face images randomly selected from the CMU PIE data set were selected, and each image was cropped to a resolution of 32×32. Further, a total of 490 images of the C05 subset are used as the source domain dataset, and a total of 490 images of the C27 subset are used as the target domain dataset to be classified.

[0097] S1. Acquire the source domain image in the source domain data set and the target domain image in the target domain data set, grayscale the source domain image and the target domain image respectively, and arrange the grayscale values ​​of the source domain image into The feature vector and the gray value of the target domain image are arranged into a feature vector. In this embodiment 2, the gray value of each image is arranged into a feature vector, that is, each image is represented as a 1024-dimensional column vector, Correspondingly, the feature matrix of the source domain data can be obtained by a...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses a cross-domain image classification method based on coupling knowledge migration, and the method comprises the steps: searching a common low-dimensional subspace of a source domain and a target domain based on a maximum mean value difference criterion, and eliminating the difference between the data edge distribution and class condition distribution of the source domain andthe target domain; constructing respective adjacent graphs according to the label information of the source domain data and the pseudo label information of the target domain data, keeping the structural consistency of the data from the original space to the low-dimensional subspace, dynamically adjusting the structures of the adjacent graphs, and promoting the positive migration of knowledge in the domain; training a nearest neighbor classifier by utilizing the source domain data with the label information in the low-dimensional subspace, and carrying out continuous iterative optimization onthe pseudo label information of the target domain data to obtain final label information of the target domain data, thereby completing cross-domain image classification; in addition, according to themethod, different confidence coefficients are given to the pseudo tags of the target domain image by designing a sample reweighting strategy, so that the negative migration of knowledge in the domainis effectively reduced, and the cross-domain image classification precision is improved.

Description

technical field [0001] The invention relates to the technical field of computer vision image classification, in particular to a cross-domain image classification method based on coupled knowledge transfer. Background technique [0002] Traditional machine learning algorithms usually require a large number of labeled data samples, and require training samples and test samples to obey independent and identical distributions. A straightforward approach is to migrate an existing labeled dataset to a new unknown dataset, i.e., from the source domain to the target domain. However, there must be differences in distribution between different data sets, and the assumption of "obeying independent and identical distribution" is often not true. Therefore, traditional image classification methods are difficult to obtain better classification performance. [0003] In order to overcome the difference in sample distribution between the source domain and the target domain and improve the g...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
IPC IPC(8): G06K9/62G06K9/46G06N3/04
CPCG06V10/40G06N3/044G06N3/045G06F18/24147
Inventor 孟敏兰孟城武继刚
Owner GUANGDONG UNIV OF TECH
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products