Segment-Based Unsupervised Learning Method in Sensor-Based Human Activity Recognition.

Area of Science:

• Computer Science
• Biomedical Engineering
• Machine Learning

Background:

• Sensor-based human activity recognition (HAR) is crucial for behavioral analysis, particularly in healthcare.
• Traditional HAR methods rely on machine learning, while deep learning offers automatic feature extraction but demands extensive labeled data.
• Labeling data for deep learning models is labor-intensive and costly.

Purpose of the Study:

• To propose a segment-based unsupervised deep learning method for HAR using accelerometer data.
• To reduce the cost of data annotation by focusing on activity segment points rather than labels.
• To develop a robust HAR system adaptable to various situations and sensor data characteristics.

Main Methods:

• Developed a segment-based unsupervised deep learning approach for HAR.
• Proposed a data collection method requiring only start, change, and end point annotations.
• Created a novel segment-based SimCLR combined with SDFD for feature representation learning.

Main Results:

• Demonstrated that the proposed combined method acquires generalized feature representations.
• Showcased the robustness of the method to sensor data sampling frequency through transfer learning.
• Validated the effectiveness of unsupervised feature learning for HAR.

Conclusions:

• The proposed segment-based unsupervised deep learning method offers an effective solution for HAR.
• The approach significantly reduces annotation costs while maintaining high performance.
• The method shows promise for real-world applications requiring adaptable and robust human activity recognition.