Dynamic Multimode Fiber Specklegram Sensor Smart Bed Enabled by Deep Learning

Fiber specklegram sensors eliminate the need for complex optical interrogators by using a camera to capture images or videos of the speckle pattern at the end facet of a multimode optical fiber. A fiber specklegram sensor is highly sensitive to movement anywhere along the length of the fiber, yielding complex information on external perturbations. The ability of the fiber specklegram sensor to be laid over a large surface area and respond to movement at any location can be useful in many real life sensing problems. Until now, only static speckle images have been used to evaluate such perturbations. However, there are applications where intrinsically dynamic measurands are desired, requiring different approaches to the analysis. In this work, we utilized deep learning techniques to analyze dynamic speckle videos for applications in dynamic biomechanical sensing, which we demonstrated for respiration rate monitoring as a proof-of-concept. This application demands both wide area sensing, that is, coverage of a mattress, together with dynamic information. We achieve this by recording videos of speckle from a single multimode fiber that covers a mattress in an S-configuration, and train convolutional neural networks directly on the video data. We show that the fiber specklegram sensor combined with our deep learning model can accurately classify the respiration rate. This approach has wide reaching potential for other biomechanical healthcare applications, such as pressure sore prevention and continuous monitoring to reduce the risk of falls.