«上一篇/Previous Article|本期目录/Table of Contents|下一篇/Next Article»

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

31

期数:

2021年06期

页码:

175-180

栏目:

应用前沿与综合

出版日期:

2021-06-10

文章信息/Info

Title:

Image Recognition Based on Generative Adversarial Network

文章编号:

1673-629X(2021)06-0175-06

作者:

程换新; 张志浩; 刘文翰; 郭占广

青岛科技大学 自动化学院,山东 青岛 265200

Author(s):

CHENG Huan-xin; ZHANG Zhi-hao; LIU Wen-han; GUO Zhan-guang

Qingdao University of Science and Technology,Qingdao 265200,China

关键词:

自然语言处理; 生成对抗网络; 深度学习; 图像识别; 准确性

Keywords:

natural language processing; generative adversarial network; deep learning; image recognition; accuracy

分类号:

TP391

DOI:

10. 3969 / j. issn. 1673-629X. 2021. 06. 031

摘要:

随着深度学习的迅速发展,图像识别技术也在日益提高。 但在日常的人脸识别、物体识别的应用中常有识别内容错误、识别准确率过低的问题。 对此,提出了一种基于生成对抗网络的图像问答模型( GAN-QA) 。 首先生成对抗网络显示出了强大的图像识别能力,通过生成对抗网络的生成器、判别器原理可以更好地提取图像特征,显著提高了图像识别的准确率。 同时根据视觉识别的自然语言处理( NLP) 也取得了极大的提升。 该模型通过长短期记忆网络(LSTM) 将两者结合起来,通过生成对抗网络识别图像,而后问题和视觉信息被输入到长短期记忆网络中,通过模型的训练可以对图像上的问题给出答案。 在数据集 DAQUQR 上的验证结果表明,所提出的基于生成对抗网络的图像问答模型能够有效地提高对带问题图像的识别问答能力,由此明显提升了图像问答的准确度。

Abstract:

With the rapid development of deep learning,image recognition technology is also increasing. However,in the daily application of face recognition and object recognition, there are often problems of wrong recognition content and low recognition accuracy.Therefore,we propose an image question answering model ( GAN - QA) based on generative adversarial network. First of all, the generative adversarial network shows a strong image recognition capability. By generator and discriminator principle of generative adversarial network,image features can be better extracted,which significantly improves the accuracy of image recognition. At the same time,natural language processing ( NLP) based on visual recognition has also been greatly improved. The model combines the two through a long-term short-term memory network ( LSTM) . The image is recognized by generative adversarial network,and then the questions and visual information are input into the long-term and short-term memory networks. Through model training,answers to the questions on the images can be given. The verification on the data set DAQUQR shows that the proposed image question answering model based on generative adversarial networks can effectively improve the ability to identify question and answer images with questions,thereby significantly improving the accuracy of image question answering.

常用功能

更新日期/Last Update: 2021-06-10