Flexible pressure sensors based on weft–knitted fabrics for real–time body signal monitoring applications with integrated tiny convolutional neural networks

• High-performance soft/flexible pressure sensor from CNT fabrics and pyralux. • High sensitivity of 0.16 kPa −1 , very fast response/recovery time of 75 ms, and small thickness of only 0.75 mm. • TinyCNN model with a tiny size of < 3 MB can fit on embedded devices with limited resources. • Real-time monitoring of two body vital signs reached 95.2% accuracy when tracking the chest movements and 88.9% accuracy when tracking the pharynx movements. In contemporary society, balancing demanding work schedules with personal health care has become increasingly challenging, often resulting in considerable psychological stress. Consequently, personal health monitoring models have emerged as a viable solution and have gained significant traction in recent years. These models are widely regarded as suitable approaches for addressing the complexities of modern life. Systems comprising compact, wearable sensor components integrated with artificial intelligence represent a promising research direction that continues to attract substantial attention from both scientific communities and startup enterprises. Aligned with this development, we propose an integrated system that combines soft, flexible sensors with embedded artificial intelligence for real-time monitoring of vital physiological parameters. The system incorporates flexible pressure sensors designed to capture respiratory pressure from the chest as well as movements of the pharyngeal region. These sensors are primarily fabricated using conductive weft–knitted fabric and flexible printed circuit boards, which enhance their sensitivity, accuracy, and user comfort, while mitigating the discomfort often associated with conventional rigid sensors. Collected signals are processed through an embedded convolutional neural network (CNN) model capable of monitoring bodily conditions and accurately classifying various physiological movements. The proposed sensors exhibit a high sensitivity of approximately 0.16 kPa −1 and a rapid response time of around 75 ms. Experimental results demonstrate that the tinyCNN model achieves classification accuracies of up to 95.2 % for respiratory signals from the chest and 88.9 % for movement signals from the pharynx in real–time conditions. This research is anticipated to contribute significantly to the field of personalized healthcare by advancing the development of intelligent and flexible sensing systems. Furthermore, the simplicity of the fabrication process enhances the potential for widespread adoption, making the proposed system a valuable and accessible reference for future wearable health monitoring technologies.