Image Enhancement Method Based on Improved DCGAN for Limit Sample

In order to solve the problem of limit sample in the detection of surface coating defects of plastic parts, an improved image dataset enhancement algorithm of Deep Convolutional Generative Adversarial Networks (DCGAN) is proposed, which is based on the DCGAN.The algorithm improves the existing activ...

Full description

Saved in:
Bibliographic Details
Main Authors: Yin, Xuehong, Hou, Bingxiang, Huang, Yong, Li, Chenyang, Fan, Zhiqiang, Liu, Jihong
Format: Conference Proceeding
Language:English
Subjects:
Explosions
Feature extraction
Generative adversarial networks
Generators
Image Dataset Enhancement
Improved DCGAN
Limit Sample
Mechatronics
Plastics
Training
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In order to solve the problem of limit sample in the detection of surface coating defects of plastic parts, an improved image dataset enhancement algorithm of Deep Convolutional Generative Adversarial Networks (DCGAN) is proposed, which is based on the DCGAN.The algorithm improves the existing activation function without changing the computational effort, and enhances the richness of the generated features, alleviating the gradient disappearance or explosion during the training process of the network. Next, introduces the deep convolutional residual blocks into the existing generator structure to obtain relatively high-resolution generated images, compares the results with Generative Adversarial Networks(GAN) and DCGAN. The results show that the improved DCGAN has a more significant improvement in the effect of image sample generation for surface painting of plastic parts.
ISSN:2157-1481
DOI:10.1109/ICMTMA54903.2022.00078