基于图卷积神经网络的WSN零动态攻击检测方法

WSN Zero Dynamic Attack Detection Method Based on Graph Convolutional Neural Network

零动态攻击与一般攻击方式相比,隐蔽性更强,因此更不容易被发现。以往常规的检测方法在检测这种攻击方式时,漏检率和误检率较高。针对上述问题,研究一种基于图卷积神经网络的WSN零动态攻击检测方法。基于零动态攻击原理,以信道状态信息作为采集源,利用CSI-Tools工具实现CSI数据包采集。从CSI数据包中分离出幅值数据和相位数据,针对前者实施去噪处理,针对后者实施校准处理。从幅值数据和相位数据中提取4个特征,以特征为输入,构建图结构,利用图卷积神经网络实现无线传感网络零动态攻击检测。结果表明:基于图卷积神经网络的攻击检测方法的漏检率和误检率相对更低,由此说明该方法对零动态攻击检测更为有效,能够实现更为准确的检测。

Compared with general attack methods,zero dynamic attacks are of stronger concealment and are therefore less likely to be detected.In the past,conventional detection methods showed high rates of missed detection and false detection when detecting attacks of this type.To address the above problems,a zero dynamic attack detection method for wireless sensor networks based on graph neural networks is studied.Based on the principle of zero dynamic attacks,channel state information is collected and CSI data packets are collected using CSI-Tools.From CSI data packets are separated amplitude data and phase data;denosing is performed on the former and calibration on the latter.From amplitude and phase data are extracted four features.They are used as inputs to construct a graph structure and graph neural networks are utilized to achieve zero dynamic attack detection in wireless sensor networks.The results show that the attack detection method based on graph neural networks has relatively lower rates of missed detection and false detection,indicating that this method is more effective in detecting zero dynamic attacks and can achieve greater accuracy of detection.