Contactless Detection of Abnormal Breathing Using Orthogonal Frequency Division Multiplexing Signals and Deep...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Contactless detection of abnormal respiratory patterns is difficult, especially with multiple individuals present.
• Software-Defined Radio (SDR) shows potential for capturing subtle respiratory movements, but multi-person interference complicates analysis.
• Distinguishing individual breathing patterns in close proximity or with similar conditions remains a challenge.

Purpose of the Study:

• To develop a contactless, non-invasive system for monitoring and classifying abnormal breathing patterns.
• To address the challenges of multi-person scenarios in respiratory monitoring.
• To achieve real-time classification of diverse respiratory conditions using advanced signal processing and deep learning.

Main Methods:

• Utilized orthogonal frequency division multiplexing (OFDM) signals captured by SDR technology.
• Developed a hybrid deep learning model (VGG16-GRU) combining CNNs and GRUs for spatial-temporal feature extraction.
• Collected a dataset in an office environment, including complex multi-subject scenarios.

Main Results:

• The system accurately detected and classified 11 distinct respiratory patterns (e.g., whooping cough, eupnea, Cheyne-Stokes, OSA).
• Achieved high performance metrics: 99.07% overall accuracy, 99.08% precision, 99.09% recall, and 99.07% F1-score.
• Demonstrated effectiveness in distinguishing individual breathing patterns in multi-person settings.

Conclusions:

• This research provides a reliable and scalable solution for contactless, real-time respiratory condition detection and classification.
• The system has significant potential for automated diagnostic tools in clinical and healthcare settings.
• Future work includes dataset expansion and model refinement for diverse real-world respiratory data.