【摘要】 针对使用视频监控的入侵检测存在的图像存储占用空间大、监测死角的问题，提出一种基于信道状态信息(Channel State Information, CSI)的入侵检测模型。该模型融合了双向长短期记忆(Bidirectional Long Short-Term Memory, BiLSTM)网络、卷积递归混合网络(Convolutional Recurrent Neural Network, CRNN)和多头注意力机制(Multi-Head Attention, MHA),结合空间特征提取、时序依赖信息的建模和全局注意力机制，增强从CSI数据中挖掘特征的能力。将处理后的CSI数据输入所提模型中进行判识，并在不同场景下对该模型进行实验测试。实验结果表明，将判别的入侵检测结果与使用长短期记忆模型(Long Short-Term Memory, LSTM)的检测结果对比，所提模型面向人员入侵检测任务的识别准确率为98%,优于LSTM模型，能够满足不同场景下的入侵检测需求。更多还原

【Abstract】 For the challenges of large image storage and surveillance blind spots in intrusion detection using video monitoring, a channel state information(CSI) based intrusion detection model is proposed.The model integrates the bidirectional long short-term memory(BiLSTM) network, the convolutional recurrent neural network(CRNN),and the multi-head attention(MHA) mechanism, combines the spatial feature extraction, the temporal dependency modeling, and the global attention mechanisms to enhance the capability of extracting features from CSI data.The processed CSI data are input into the proposed model for identification, and experiments are conducted in various scenarios to test the model.The intrusion detection results are compared with the detection results obtained from the long short-term memory(LSTM) model.Experiment results show that the proposed model achieves an accuracy of 98% for personnel intrusion detection tasks, outperforms the LSTM model, and can meet the intrusion detection requirements in different scenarios.更多还原