Modern Cough Detection and Classification System Using Machine Learning and IoT
The research team developed an innovative system combining machine learning (ML) techniques with Internet of Things (IoT) technologies to automatically detect and classify coughs in real-time. This interdisciplinary solution marks a breakthrough in remote health monitoring, particularly in the context of respiratory diseases such as asthma, COPD, and viral infections, including COVID-19.
At the core of the system is advanced acoustic signal analysis using deep learning models. Modern neural network architectures, such as MobileNet, RESNET-50, and classical convolutional neural networks (CNN), were used to classify cough sounds. Among the tested models, MobileNet achieved the highest performance, with an accuracy of 84%, an AUC value of 0.902, and an F1-score of 0.846. Such a high level of classification precision allows not only the detection of a cough but also the differentiation of its types, which is crucial for differential diagnostics.
The system uses the analysis of sound signal spectrograms, which enables the identification of unique acoustic features for different types of coughs. This allows for the distinction, for example, between dry and wet coughs, which in clinical practice can assist doctors in diagnosing the course of an infection and adjusting therapy accordingly.
An important element of the designed solution is the protection of privacy and the security of medical data. All signal processing and classification operations are performed locally on the Jetson Nano platform, eliminating the need to transmit audio data to the cloud. Communication with external systems is carried out via the lightweight and secure MQTT protocol, ensuring efficient and confidential transmission of the analysis results.
The system also offers real-time remote monitoring functionality – the classification results are immediately available to doctors and patients through a mobile application and web panel. This solution can significantly streamline diagnostic and therapeutic processes, especially in the care of patients with chronic diseases or in areas with limited access to medical facilities.
The implementation of this technology opens new opportunities in preventive medicine, home care, and telemedicine. Automatic cough detection can be used not only in monitoring individual cases but also in scalable early warning systems in healthcare facilities and public spaces.
The full version of the scientific publication is available at:
link to publication