摘要
卷积神经网络CNN目前在有监督学习领域有着优秀的表现,但在无监督学习领域研究进展缓慢。该文将CNN引入到GAN中的生成式模型中做无监督训练,利用CNN强大的特征提取能力来提高生成式模型的学习效果,采用TensorFlow和Python代码实现了DCGAN中的D模型和G模型,并在MNIST部分数据集下验证了模型生成数字图像效果。实验结果表明采用DCGAN可以有效获取图像表征用于分类并具备生成较高分辨率的图像能力。
Convolutional neural network(CNN) has excellent performance in supervised learning field, but the progress of research is slow in unsupervised learning. This paper introduced CNN into the generative model of GAN in unsupervised training,to improve the ability to extract generative model learning effect by the powerful character of CNN, using Tensor Flow and Python code to achieve the D model and G model within DCGAN, and validates the model to generate digital image effect by the dataset of MNIST. Experimental results show that DCGAN can effectively obtain image representation for classification, and has the ability to generate high resolution of images.
出处
《电脑知识与技术》
2017年第12X期219-221,共3页
Computer Knowledge and Technology