摘要
目前人工智能技术已经渗透到媒体日常生产的全部环节,研究面向内容安全的图像识别技术,对网络中的违法不良图像信息进行识别和监管,具有重要的现实意义。本设计利用深度可分离卷积神经网络和MobileNet模型,配合cuDNN的GPU并行计算架构,对构建的敏感图像数据集进行训练。保存训练完成的模型,对敏感图像实现较高准确度的识别。
At present,the artificial intelligence technology has penetrated into all aspects of daily media production.It has important realistic significance to research image recognition technology oriented content security,identify and regulate illegal information of the network.This paper uses deep separable convolution neural network,MobileNet model,GPU parallel computing architecture in combination with cuDNN to train the constructed sensitive image data sets.After saving the trained model,a high accurate recognition to sensitive images is realized.
作者
邢艳芳
卓文鑫
段红秀
XING Yanfang;ZHUO Wenxin;DUAN Hongxiu(Nanguang College of Communication University of China,Nanjing 211172,China)
出处
《电视技术》
2018年第7期53-56,共4页
Video Engineering
基金
江苏省高等学校自然科学研究面上资助项目(17KJB510054)--人工智能在传媒领域内容监管中的应用研究
