基于Swin Transformer的暴力敏感图像分类研究

Research on violence sensitive image classification based on Swin Transformer

随着互联网的普及,网络社交媒体平台的用户数量呈爆炸式增长。同时,相关平台上的音视频数据不断增多,其中的暴力敏感图像给公众的身心健康带来了极大危害。在此情况下,各大网络平台的内容安全审核工作越来越繁重,传统的人工审核方式显然无法满足实际需求。深度学习技术的发展为以上问题带来了新的解决方案,可对暴力敏感图像进行智能化识别。文章基于Swin Transformer模型进行了相关改进,构建了用于暴力敏感图像分类的SESTNet模型,并在综合数据集中对其识别和分类能力进行了验证。

With the popularization of the Internet,the number of users of the online social media platform has exploded.At the same time,the increasing amount of audio and video data on relevant platforms,including violent and sensitive images,poses great harm to the physical and mental health of the public.In this situation,the content security review work of major online platforms is becoming increasingly heavy,and traditional manual review methods are obviously unable to meet practical needs.The development of deep learning technology has brought new solutions to the above problems,which can intelligently recognize violence sensitive images.The article made relevant improvements based on the Swin Transformer model,constructed the SESTNet model for violence sensitive image classification,and verified its recognition and classification capabilities on a comprehensive dataset.