%0 Journal Article
%T 基于改进储备池计算模型的人体行为识别<br>Human Activity Recognition Based on Enhanced Reservoir Computing
%A 陈宇骞
%A 任旭超
%A 朱雅琳
%A 颜延
%J Computer Science and Application
%P 528-536
%@ 2161-881X
%D 2023
%I Hans Publishing
%R 10.12677/CSA.2023.133052
%X <span style=\"font-size:14px;\">人体行为识别(human activity recognition, HAR)是元宇宙场景人机交互应用中的关键任务。在基于传感器的HAR任务中，提取有用特征是提高模型性能的关键。为此，本文提出了一种改进的储备池计算网络框架(enhanced reservoir computing network framework, ERCNF)。本文首先使用圆形储备池模块提取信号的特征，然后使用降维模块对提取的特征进行降维，最后使用岭回归器进行线性读出。我们在UCI-HAR和mHealth两个公开数据集上评估ERCNF模型。在UCI-HAR和mHealth数据集上ERCNF模型的准确率分别为98.1%和95.6%。该模型具有训练速度快，准确率高等特点，能有效地从数据中提取深度特征，在可穿戴应用中具有良好的应用前景。</span><br />
<span style=\"font-size:14px;\">Human activity recognition (HAR) is a critical task in human-machine interaction applications for meta-universe scenarios. Extracting useful features are the keys to improving the model performance in a sensor-based HAR tasks. This paper proposes an enhanced reservoir computing network framework (ERCNF) to address the problems. In this work, the ERCNF model is feature extracted by the circular reservoir topology module, then the extracted features are dimensionally reduced by a dimensionality reduction module, and finally linear readout is performed by a ridge regressor. We evaluate the ERCNF model with two benchmark HAR on the datasets of UCI-HAR and mHealth. The best classification accuracies achieved by the ERCNF model are 98.1% and 95.6% with UCI-HAR and mHealth datasets, respectively. The model has fast training speed and high accuracy, and can effectively extract depth features from the data, which has good application prospects in wearable ap-plications.</span>
%K 储备池计算，回声状态网络，特征融合，人体行为识别<br>Reservoir Computing (RC)
%K Echo State Network
%K Feature Fusion
%K Human Activity Recognition
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=63519