Contactless Detection of Abnormal Breathing Using Orthogonal Frequency Division Multiplexing Signals and Deep Learning in Multi-Person Scenarios.

Contactless detection and classification of abnormal respiratory patterns is challenging, especially in multi-person scenarios. While Software-Defined Radio (SDR) systems have shown promise in capturing subtle respiratory movements, the presence of multiple people introduces interference and complexity, making it difficult to distinguish individual breathing patterns, particularly when subjects are close together or have similar respiratory conditions. This paper presents a contactless, non-invasive system for monitoring and classifying abnormal breathing patterns in both single and multi-person scenarios using orthogonal frequency division multiplexing (OFDM) signals and deep learning techniques. The system automatically detects various respiratory patterns, such as whooping cough, Acute Cough, eupnea, Bradypnea, tachypnea, Biot's, sighing, Cheyne-Stokes, Kussmaul, CSA, and OSA. Using SDR technology, the system leverages OFDM signals to detect subtle respiratory movements, allowing real-time classification in different environments. A hybrid deep learning model, VGG16-GRU, combining convolutional neural networks (CNNs) and gated recurrent units (GRUs), was developed to capture both spatial and temporal features of continuous respiratory data. The model successfully classified 11 distinct breathing patterns with high accuracy, achieving an overall accuracy of 99.07%, precision of 99.08%, recall of 99.09%, and an F1-score of 99.07%. The dataset, collected in an office environment, includes complex scenarios with multiple subjects, demonstrating the system's effectiveness in distinguishing individual breathing patterns, even in multi-person settings. This research advances contactless respiratory monitoring by offering a reliable, scalable solution for real-time detection and classification of respiratory conditions. It has significant implications for the development of automated diagnostic tools for respiratory disorders, offering potential benefits for clinical and healthcare applications. Future work will expand the dataset and refine the models to improve performance across diverse respiratory patterns and real-world data from a respiratory unit.