SlideAugment: A Simple Data Processing Method to Enhance Human Activity Recognition Accuracy Based on WiFi.

Area of Science:

• Computer Science
• Signal Processing

Background:

• Activity recognition using WiFi signals is an active research area.
• Existing datasets often lack sufficient data for robust model training.
• Channel State Information (CSI) is a key feature for WiFi-based sensing.

Purpose of the Study:

• To introduce a novel data augmentation technique for WiFi-based activity recognition.
• To address the challenge of insufficient data in current datasets.
• To improve the accuracy and generalizability of activity recognition models.

Main Methods:

• Proposed a data augmentation method named window slicing.
• Window slicing generates multiple samples from a single raw data point.
• Applied the method to both a public dataset and a newly collected dataset.

Main Results:

• Achieved a significant improvement in activity recognition accuracy on a public dataset, increasing from 88.13% to 97.12%.
• Demonstrated accuracy improvements on the collected dataset as well.
• The method proved effective in enhancing recognition performance.

Conclusions:

• Window slicing is a simple yet effective general-purpose data augmentation technique for CSI data.
• The proposed method enhances WiFi-based activity recognition accuracy.
• The technique offers good interpretability and applicability to various datasets.