深度模型集成的不良图像分类

移动通信技术的飞速发展在提升用户通信体验的同时也为不良信息的散布提供了便利,针对如何在大量数据中进行不良内容的识别与过滤问题,提出一种基于深度模型集成的不良图像分类模型(EDM),通过集成多个结构不同、信息互补的深度模型来最优地区分分布差异较大的不良图像.为了验证本方法的有效性,建立一个真实移动通信场景下的不良图像数据集,并在此数据集上与基于传统支持向量机(SVM)的不良图像分类模型、基于深度卷积神经网络的Alexnet、VGG与Googlenet分类模型做对比.实验结果表明:本文所提深度模型集成方法在不良图像分类性能上明显优于其他模型,分类精度、精确率和召回率分别达到94%、84%和98%.

Illegal image classification based on ensemble deep model

The rapid development of mobile communication technology has greatly promoted the communication experience of users.How to identify and filter out the illegal content in a large a-mount of data is crucial for improving the ability and level of illegal information management in China Mobile.Towards this end,this paper proposes an ensemble deep model (EDM)to classify illegal images.In this approach,several deep models with diverse network structures and comple-mentary information are integrated by using the proposed scheme,and the illegal images with di-verse distributions will be distinguished.To evaluate the effectiveness of the proposed approach, we first collect and set up an illegal image dataset,and compare the proposed approach with the traditional support vector machine(SVM)based image classification method and Alexnet-based, VGG-based and Googlenet-based methods.Experiments show that the proposed approach clearly outperforms the existing methods and obtains excellent classification performance in accurate (94％),precision (84％)and recall (98％).