Scholarlink

<p>Human activity recognition (HAR) has become a prevalent research topic in Artificial Intelligence. Recently, numerous deep learning (DL) methods have been extensively applied to HAR, due to their effective automatic feature extraction abilities and to tackle HAR challenges like time-series raw data. In this paper, a Convolutional Neural Network (CNN) integrated with Stacked Long Short-Term Memory (SLSTM) is proposed to predict human activity recognition by extracting spatial and temporal features from the HAR data. The gathered raw data from portable sensors is first processed through a deep neural network that includes two convolutional layers, followed by two LSTM layers whose output is then passed to a dense layer, followed by a fully connected layer with a Softmax activation function to classify input data. The accuracy of the suggested model is 99.86% and 94.44% on the PAMAP2 and UCI-HAR datasets, respectively. The offered network shows improvements on both temporal and spatial dimensions, enhancing the recognition rate when testing the two datasets. The accuracy is substantially enhanced associated with state-of-the-art methods.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

CNN-SLSTM framework for human activity recognition using wearable sensor data

  • 期刊论文

  • 发表日期：

    2025年6月26日

查看全文
  • P. Krishnaleela,
  • R. Meena Prakash

摘要

Human activity recognition (HAR) has become a prevalent research topic in Artificial Intelligence. Recently, numerous deep learning (DL) methods have been extensively applied to HAR, due to their effective automatic feature extraction abilities and to tackle HAR challenges like time-series raw data. In this paper, a Convolutional Neural Network (CNN) integrated with Stacked Long Short-Term Memory (SLSTM) is proposed to predict human activity recognition by extracting spatial and temporal features from the HAR data. The gathered raw data from portable sensors is first processed through a deep neural network that includes two convolutional layers, followed by two LSTM layers whose output is then passed to a dense layer, followed by a fully connected layer with a Softmax activation function to classify input data. The accuracy of the suggested model is 99.86% and 94.44% on the PAMAP2 and UCI-HAR datasets, respectively. The offered network shows improvements on both temporal and spatial dimensions, enhancing the recognition rate when testing the two datasets. The accuracy is substantially enhanced associated with state-of-the-art methods.