【摘要】 实时检测电力作业违章行为是保障电力企业安全生产的一个重要环节。然而，已有的视频监控、无人机航拍、穿戴设备监控等方法难以准确捕捉和识别不可见电力作业场景下的习惯性或故意违章行为。该文提出一种基于Wi-Fi信道状态信息(channel state information, CSI)的电力作业违规动作识别模型，该模型首先通过格拉姆矩阵分割WIFI CSI无线信号，以获取人体动作的空间局部特征和全局特征；接着，通过计算中位数绝对偏差均值，捕获不受环境和身体姿态影响的人体动作动态分量；进而，提取CSI信号中的动作时序特征；然后，引入注意力机制实现多尺度空间特征和时序特征的融合；最后，利用公开的人体动作感知数据集及自建的电力操作场景数据集，与5种成熟的无线感知算法进行了对比分析。实验结果表明，提出的方法能够有效消除环境因素影响，违章行为识别准确率较高。更多还原

【Abstract】 Real-time detection of power operation violations is a crucial aspect in ensuring the safety production of power enterprises.However, existing methods such as video surveillance, unmanned aerial vehicle(UAV) aerial photography, and wearable device monitoring struggle to accurately capture and identify habitual or intentional violations in invisible power operation scenarios.This paper proposes a Wi-Fi channel state information(CSI)-based model for recognizing power operation violations.The model first segments Wi-Fi CSI wireless signals using Gramian matrices to extract spatial local and global features of human movements; subsequently, it captures dynamic components of human actions unaffected by environmental factors or body postures by calculating the mean median absolute deviation; further, temporal features of actions within CSI signals are extracted; then, an attention mechanism is introduced to fuse multi-scale spatial and temporal features; finally, comparative analysis is conducted against five mature wireless sensing algorithms using publicly available human activity recognition datasets and self-constructed power operation scenario datasets.Experimental results demonstrate that the proposed method effectively mitigates environmental influences while achieving high accuracy in identifying violation behaviors.更多还原