Wyniki wyszukiwania - BazTech

1

A multi-scale time segments attention mechanism for sensor-based human activity recognition

Rong Hailong, Wang Hao, Wu Xiaohui, Jin Tianlei, Zou Ling

Metrology and Measurement Systems

|

2025

|

Vol. 32, nr 1

1--17

EN

Sensor-based Human Activity Recognition (SHAR) technology is dedicated to utilizing sensor signals from smart devices to detect and identify human activities, thereby assisting in daily life. With the successful application of deep learning techniques, researchers are exploring the potential of integrating them with SHAR. Traditional fixed sliding window methods for processing datasets often lead to multi-class activity mixing. To alleviate this issue, researchers have introduced time attention mechanisms to focus on key temporal points related to activities. To address this challenge, we propose an innovative Multi-scale Time Segments Attention Mechanism (MTSA), which diverges from traditional time attention mechanisms by focusing on time segments pertinent to activities, better aligning with the characteristics of SHAR data and significantly reducing computational resource consumption. Our experiments on recognized datasets such as UCI-HAR, PAMAP2, and WISDM validate the effectiveness of MTSA, demonstrating that it can be seamlessly integrated into existing SHAR models, enhancing performance without adding extra computational overhead.

2

Attitude estimation based on multi-scale grouped spatio-temporal attention neural networks

Rong Hailong, Wu Xiaohui, Wang Hao, Jin Tianlei, Zou Ling

Metrology and Measurement Systems

|

2024

|

Vol. 31, nr 1

195--211

EN

In recent years, due to the proliferation of inertial measurement units (IMUs) in mobile devices such as smartphones, attitude estimation using inertial and magnetic sensors has been the subject of considerable research. Traditional methods involve probabilistic and iterative state estimation; however, these approaches do not generalize well over continuously changing motion dynamics and environmental conditions. Therefore, this paper proposes a deep learning-based approach for attitude estimation. This approach segments data from sensors into different windows and estimates attitude by separately extracting local features and global features from sensor data using a residual network (ResNet18) and a long short-term memory network (LSTM). To improve the accuracy of attitude estimation, a multi-scale attention mechanism is designed within ResNet18 to capture finer temporal information in the sensor data. The experimental results indicate that the accuracy of attitude estimation using this method surpasses that of other methods proposed in recent years.

3

A hybrid denoising method for gyroscopes based on multiple screening

Rong Hailong, Jin Tianlei, Wang Hao, Wu Xiaohui, Zou Ling

Metrology and Measurement Systems

|

2023

|

Vol. 30, nr 4

755--772

EN

To reduce the random error of microelectromechanical system (MEMS) gyroscope, a hybrid method combining improved empirical mode decomposition (EMD) and least squares algorithm (LS) is proposed. Firstly, based on the multiple screening mechanism, intrinsic mode functions (IMFs) from the first decomposition are divided into noise IMFs, strong noise mixed IMFs, weak noise mixed IMFs and signal IMFs. Secondly, according to their characteristics, they are processed again. IMFs from the second decomposition are divided into noise IMFs and signal IMFs. Finally, useful signal is gathered to obtain the final denoising signal. Compared with some other denoising methods proposed in recent years, the experimental results show that the proposed method has obvious advantages in suppressing random error, greatly improving the signal quality and improving the accuracy of inertial navigation.