A Multitask Deep Learning Approach for Sensor-Based Human Activity Recognition and Segmentation
A Multitask Deep Learning Approach for Sensor-Based Human Activity Recognition and Segmentation
复制标题
DOI：
10.1109/tim.2023.3273673
复制
发表时间：
2023-03
影响因子：
5.6
通讯作者：
Furong Duan;Tao Zhu;Jinqiang Wang;L. Chen;Huansheng Ning;Yaping Wan
中科院分区：
文献类型：
--
作者：
Furong Duan;Tao Zhu;Jinqiang Wang;L. Chen;Huansheng Ning;Yaping Wan
基于传感器的人类活动识别（HAR）的深度学习（DL）是近年来的研究热点。传感器数据流分割是HAR的核心内容，目前被视为一个独立的预处理任务，通常具有固定大小的窗口。这导致了两个关键问题，即在一个固定大小的窗口内可能有多个活动引起的多类窗口问题，以及由于噪声数据和过度分割引起的预测结果波动。为了解决这些研究挑战，在这篇文章中，我们设想了一种新的多任务DL方法，同时分割和识别人类活动。具体来说，我们提出了一种基于特征序列生成的多尺度窗口方法，以克服多类窗口问题。我们开发了一种新的边界偏移预测算法来调整窗口的边界，以解决过分割问题。此外，我们设计了一个多任务框架，以简化和优化的活动识别和分割任务同时进行。我们在八个基准数据集上进行了广泛的实验，以评估所提出的框架和相关方法。初步结果表明，我们的方法优于目前国家的最先进的HAR方法的性能。
Deep learning (DL) for sensor-based human activity recognition (HAR) has been a focus of research in recent years. Sensor data stream segmentation is a core element in HAR, which has currently been treated as an independent preprocessing task, usually with a fixed-size window. This has led to two critical problems, namely the multiclass window problem caused by possible multiple activities within a fixed-size window and the fluctuation of prediction results due to noisy data and oversegmentation. To address these research challenges, in this article, we conceive a novel multitask DL approach to segmenting and recognizing human activity simultaneously. Specifically, we propose a multiscale window method based on feature sequence generation to overcome the multiclass window problem. We develop a novel boundary offset prediction algorithm to adjust a window’s boundary to tackle the oversegmentation issue. In addition, we design a multitask framework to streamline and optimize the activity recognition and segmentation tasks simultaneously. We conduct extensive experiments on eight benchmark datasets to evaluate the proposed framework and associated methods. Initial results show that our approach outperforms the performance of current state-of-the-art HAR methods.