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

Classification method based on Triple-GAN

A classification method and classifier technology, applied in the field of image processing, can solve problems such as gradient disappearance, uneven labeling, unstable model training, etc., and achieve the effect of strong adaptability

Active Publication Date: 2018-09-11
XIANGTAN UNIV
View PDF4 Cites 13 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] However, there are still the following problems in the application process of Triple-GAN: (1) Due to the use of KL divergence distribution, for the case where the distribution does not cross, it is easy to cause the gradient to disappear, which will cause the training to terminate when the training does not achieve the desired result, thus Make model training unstable
(2) Due to the uneven labeling caused by manual labeling of samples, there is a problem that the workload of labeling a large amount of data is too large

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
  • Classification method based on Triple-GAN
  • Classification method based on Triple-GAN
  • Classification method based on Triple-GAN

Examples

Experimental program
Comparison scheme
Effect test

Embodiment

[0101] Now widely used dataset MNIST. Among them, MNIST includes 60,000 training samples, 10,000 validation samples and 10,000 test samples of handwritten digits with a size of 28×28.

[0102] (1) The MNIST dataset is available on Kaggle. After downloading the train.csv to the data / folder, load the file into the training data.

[0103] (2) The generator generates sample data marked with pseudo-labels according to the training data.

[0104] Input the 10,000 verification sample data set of MNIST into the classifier (C), establish a decision tree with a depth of M according to the random forest algorithm based on the binary decision tree, set the height of the decision tree to 5, and establish 304 binary decisions Trees, respectively built binary decision tree label labels.

[0105] (3) The classifier predicts the joint distribution of samples and class labels:

[0106] 60,000 MNIST training data sets are input into the generator, and the generated data x and class label y ...

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

Generative adversarial networks (GAN) display a great development prospect in image generation and semi-supervised learning, and have been developed to Triple-GAN. However, there are two problems requiring solving based on the classification method based on Triple-GAN, wherein the two problems comprise a problem that KL-based divergence distribution construction is easy to generate gradient disappearance to cause unstable training and a problem that the Triple-GAN employs a manual mode to perform mark of tags for samples to cause too large workload of manual marking and non-uniform marking. Based on this conditions, RandomForests are employed to perform classification of real samples to perform automatic marking of tags of blade nodes, and the ideal of the least square generative adversarial networks (LSGAN) are employed to construct a loss function to avoid gradient disappearance.

Description

technical field [0001] The invention relates to an image classification method, in particular to a method based on The invention relates to a method for fast classification of images combining distribution, Triple-GAN three-person game and random forest, and belongs to the field of image processing. Background technique [0002] Image mining is an emerging field in data mining. Image classification is the basis of data mining, facing a large amount of image data, image classification becomes more and more important. There are many classification techniques for image classification, such as decision tree, minimum distance method, neural network, fuzzy classification, support vector machine, k-means, etc. The proposal of the GAN model has raised the field of image classification to a new height, and also promoted the development of image mining technology. The research based on the GAN model has also become a research hotspot. [0003] Generative confrontation network (GA...

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): G06K9/62
CPCG06F18/24323G06F18/214
Inventor 欧阳建权方昆唐欢容
Owner XIANGTAN UNIV
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

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com