摘要
针对传统深度卷积生成网络收敛速度慢、稳定性较差的问题,本文在传统深度卷积生成对抗网络的基础上,提出了深度卷积生成对抗网络的优化算法。首先在预处理部分,融合了Canny算子和Prewitt算子的多个方向的卷积核来初始化输入图片参数,同时训练模块。为了减少训练时间,将训练分为3个阶段,每个阶段都采用不同的损失函数,从而提升网络的收敛速度及识别效果。最后再将训练后的判别网络中的卷积神经网络用来提取图像特征。LFW和CIFAR-100的实验证明,本文提出的算法具有很高的可行性和有效性,比传统生成对抗网络、CNN等图像识别具有更高的识别成功率,达到89.5%,为生成对抗网络在计算机视觉领域的应用提供了有益的参考。
In view of the slow convergence and poor stability of traditional deep convolutional generation networks,this paper combines the Connaught kernels of multiple directions of Canny operator and Prewitt operator on the basis of traditional deep convolution generation confrontation network.To initialize the input picture parameters and reduce the training time,the training is divided into three stages,each stage adopts a different loss function,thereby improving the convergence speed and recognition effect of the network and then discriminating the volume in the network after training.The neural network is used to extract image features.The experiments of LFW and CIFAR-100 prove that the proposed algorithm is highly feasible and effective,and provides a useful reference for generating anti-network applications in the field of image recognition.
作者
刘恋秋
LIU Lian-qiu(Chongqing College of Finance and Economics,Chongqing 402160,China)
出处
《液晶与显示》
CAS
CSCD
北大核心
2020年第4期383-388,共6页
Chinese Journal of Liquid Crystals and Displays
关键词
生成对抗网络
深度卷积
特征提取
计算机视觉
generate confrontation network
deep convolution
feature extraction
computer vision
