基于CSI与Attention-BiLSTM的动作识别算法

Action Recognition Algorithm Based on CSI and Attention-BiLSTM

与传统动作识别技术相比,基于信道状态信息的动作识别具有成本低、安全便利等特点,应用前景广阔。利用乐鑫ESP32采集信道子载波幅值信息,结合预处理算法,并基于结合注意力机制的双向长短期记忆网络的动作识别算法,实现对走路、拖地、捡起、坐下、蹲下和站起六种动作的特征提取与分类识别。测试结果表明:算法在测试集上的平均识别准确率高达95.8%,相较于常规的长短期记忆算法,识别准确率更高、收敛速度更快;与传统基于统计特征与机器学习的分类算法相比,算法直接利用神经网络自动提取时序特征,特征提取更精确,准确率提升超过10%。试验结果验证了该算法在基于信道状态信息的动作识别上的有效性,说明该算法具有较高的实用价值。

Compared with traditional action recognition techniques,the action recognition based on channel state information has the characteristics of low cost,safety and convenience,and has broad application prospects.Using Lexin ESP32 to collect channel subcarrier amplitude information,combined with preprocessing algorithms,and based on the action recognition algorithm of a bidirectional long short-term memory(LSTM)network combined with attention mechanism,the feature extraction and classification recognition of six types of actions,namely walking,mopping,picking up,sitting down,squatting,and standing up,were achieved.The test results show that the algorithm has the average recognition accuracy of 95.8%on the test set,which is higher in recognition accuracy and faster in convergence compared to conventional LSTM algorithms;compared with traditional classification algorithms based on statistical features and machine learning,this algorithm directly utilizes neural networks to automatically extract temporal features,resulting in more accurate feature extraction and an accuracy improvement of over 10%.The experiment result has verified the effectiveness of the algorithm in action recognition based on channel state information,indicating that the algorithm has high practical value.