War通过预处理输入数据检测对抗样本-arXiv.PDF

War：通过预处理输入数据检测对抗样本 王华 王婕 殷赵霞 （安徽大学计算机科学与技术学院 合肥 230000） Deep NeuralNetworks DNNs 摘 要：深度神经网络 （ ， ）在图像分类和语音识别等多个领域展现了其卓越 DNNs Adversarial examples 的性能。然而， 图像分类器很容易受到对抗样本 （ ）的干扰，最终导致神经网络 War WebPcompression andresize 模型得到错误的分类输出。基于此，本文提出一种基于 （ ）的检测对抗样 本的方法。该方法以WebP 压缩为核心，首先对输入图像进行WebP 压缩，再对压缩后的图像进行适当收 缩，使得对抗样本的标签发生改变，从而检测出对抗样本的存在。实验结果表明，本文所提出的方法能够 IFGSM DeepFool CW HGD 10% 有效的抵御 ， 以及 三种对抗攻击，并在识别精度上相比于 方法提升了 以 5% 上，在对抗样本检测成功率上比特征压缩方法提升 。本文方法可以有效的减少对抗样本中细小的噪声扰 动，在保证模型对原始样本识别精确度的同时根据样本标签的变化准确地检测出对抗样本。 关键词： 深度神经网络；图像分类；对抗样本；噪声扰动；WebP 压缩 War:Detectingadversarial examplesbypre-processinginput data WangHua,WangJie,YinZhaoxia (School ofComputer Science andTechnology,AnhuiUniversity,Hefei 230000) Abstract: Deep neural networks (DNNs) have demonstrated their outstanding performance in many fields such as image classification and speech recognition. However, DNNs image classifiers are susceptible to interference from adversarial examples, which ultimately leads to incorrect classification output of neural network models. Based on this, this paper proposes a methodbased on War (WebP compression and resize) to detect adversarial examples.The method takes WebP compression as the core, firstly performs WebP compression on the input image, and then appropriately resizes the compressed image, so that the label of the adversarial example changes, thereby detecting the existence of the adversarial image. The experimental results show that the proposed method