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。
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
DOI:10.19850/j.cnki.2096-4706.2021.21.026
中图分类号：TP18 文献标识码：A 文章编号：2096-4706（2021）21-0102-03
Improvement and Application of Training Technology Based on GANs
YANG Zhen, LI Dan
(Jincheng College of Sichuan University, Chengdu 611731, China)
Abstract: 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.
Keywords: GANs network's technology improvement; new dataset; network organization; feature matching; polymorphic difference learning
参考文献：
[1] YAN X C，YANG J M，SOHNK. Attribute2Image：Conditional Image Generation from Visual Attributes [J/OL].arXiv：1512.00570 [cs.LG].[2021-08-22].https://arxiv.org/abs/1512.00570.
[2] VINYALS O，TOSHEV A，BENGIO S，et al. Show and tell：A neural image caption generator [C]//2015 IEEE Conference on Computer Vision and Pattern Recognition （CVPR）.Boston：IEEE，2015，3156-3164.
[3] FUKUMIZU K，GRETTON A，SUN X H，et al. Kernel Measures of Conditional Dependence [EB/OL].[2021-08-22].http://citeseerx.ist.psu.edu/viewdoc/download；jsessionid=C6AE3A04010B73 BEE01147FB4BD2FCC0?doi=10.1.1.143.5575&rep=rep1&type=pdf.
[4] LI Y J，SWERSKY K，ZEMEL R. Generative Moment Matching Networks [J/OL].arXiv：1502.02761 [cs.LG].[2021-08-22].https://arxiv.org/abs/1502.02761.
[5] 翁邦碧，杨波，姚璞，等 . 应用多媒体与实战模拟训练法改进自救互救技术教学 [J]. 西南军医，2020，22（5）：478-481.
[6] 胡涛，李金龙.基于单阶段GANs的文本生成图像模型 [J].信息技术与网络安全，2021，40（6）：50-55.
作者简介：杨振（2000—），男，汉族，四川井研人，本科在读，研究方向：人工智能。