Human Health Activity Recognition Algorithm in Wireless Sensor Networks Based on Metric Learning

Area of Science:

• Computer Science
• Wireless Sensor Networks
• Machine Learning

Background:

• Wireless sensor networks (WSNs) are crucial for remote monitoring and data collection in complex environments.
• Human activity recognition (HAR) is a key area in computer science with applications in health and surveillance.
• Metric learning often faces challenges with small sample sizes and linear inseparability.

Purpose of the Study:

• To propose a novel human activity recognition algorithm tailored for wireless sensor networks.
• To overcome the limitations of small sample problems and linear inseparability in metric learning for HAR.
• To enhance the performance of HAR algorithms in WSNs.

Main Methods:

• The proposed algorithm utilizes kernel functions to map feature spaces into a high-dimensional, linearly separable kernel space.
• It calculates distances between samples in projected feature subspaces using two distinct distance measurement functions.
• These functions are then linearly combined with weights to form a final, robust distance measurement function.

Main Results:

• The algorithm effectively addresses the singularity of the intraclass divergence matrix, mitigating the impact of small sample issues.
• By projecting data into a high-dimensional kernel space, the method achieves linear separability for previously inseparable samples.
• This leads to improved accuracy in human activity recognition within wireless sensor network contexts.

Conclusions:

• The developed human activity recognition algorithm offers a robust solution for WSNs, particularly when dealing with limited data.
• The kernel-based approach enhances the discriminative power of the metric learning process.
• This research contributes to more reliable and accurate human activity monitoring using wireless sensor networks.