基于Wavelet-CNN网络的人类活动识别技术

Human Activity Recognition Based on Wavelet-CNN Architecture

针对传统的识别方法不能满足人类活动识别(Human Activity Recognition,HAR)技术研究需求的现状,提出了一种基于小波变换和卷积神经网络(Convolutional Neural Networks,CNN)相结合的深度学习模型。将多通道传感器的波形数据通过小波变换分解并重组作为输入。利用不同卷积核的CNN高效提取多维特征,使用最大池化层对人体无意识抖动引起的干扰噪声进行滤波操作。经过全连接层输出分类,实现对人体活动状态的准确识别。实验分别从模型收敛速度、损耗和精度三方面评估了模型性能,并在OPPORTUNITY公共数据集上与较先进的识别模型进行了对比。实验结果表明,提出的小波变化卷积网络Wavelet-CNN实现了91.65%的F1分数,具有更高的活动识别能力。

As traditional recognition methods cannot meet the research needs of Human Activity Recognition(HAR),a deep learning model based on a combination of Wavelet Transform(WT)and convolutional neural network is proposed.The waveform data of the multi-channel sensor is decomposed and reorganized through wavelet transform as input.Then,the convolutional neural network with different convolution kernels is used to efficiently extract the multi-dimensional features,and the max-pooling layers are used to filter the interference noise caused by the unconscious jitter of human body.Finally,accurate recognition of human body activity state is realized through the output classification of the fully connected layer.The model performance is evaluated by experiment in three aspects including model convergence speed,loss and accuracy,and it is compared with that of the state-of-the-art recognition models on the OPPORTUNITY public dataset.Experimental results show that the proposed Wavelet-CNN architecture achieves 91.65%F1 score and is of higher activity recognition capability.