基于GANs的训练技术的改进及应用  

Improvement and Application of Training Technology Based on GANs

作　　者：杨振 李丹 YANG Zhen;LI Dan(Jincheng College of Sichuan University,Chengdu 611731,China)

机构地区：[1]四川大学锦城学院,四川成都611731

出　　处：《现代信息科技》2021年第21期102-104,共3页Modern Information Technology

摘　　要：GANs作为典型的生成对抗网络,具有较高的应用价值和发展潜力,文章改进了GANs网络的一些新技术(训练深度生成半质量网络dcgan),针对其中部分缺点进行了有效改善,从网络架构、特征匹配和多形差值学习技术维度进行优化,并将其与bird数据集*80进行训练,得出观察结果,为未来的应用前景和环境拓展研究提供进一步可能性,以期在更多的场景中使用该算法。As a typical generation countermeasure network,GANs has higher application value and development potential.In this paper,some new technologies of GANs network are improved (training depth generation semi-mass network dcgan),and some of the shortcomings are effectively improved.The algorithm is optimized from the network architecture,feature matching and polymorphic difference learning technology dimensions,and is trained with bird dataset *80 to obtain the observation results,providing further possibility for the research of future application prospects and environment expansion,in order to use the algorithm in more scenarios.

关 键 词：GANs网络的技术改进 数据新集 网络机构 特征匹配 多形差值学习 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]