FBG-SenseGlove: A Digital Twin Data Glove for Underwater Gesture Recognition Using Deep Learning

Hand posture detection in underwater environments is a crucial method for human-computer interaction, widely applied in areas, such as diving rescue, underwater engineering, and marine scientific exploration. Vision-based gesture recognition systems are susceptible to ambient light and water turbidity, while other electrical sensing technologies often lack sufficient sensitivity and reliability under dynamic underwater conditions. To address these challenges, we designed a wearable data glove based on the fiber Bragg grating (FBG) technology, called the FBG-SenseGlove. We utilized food-grade silicone to achieve integrated encapsulation of the FBG sensors within the glove. In strain tests, the sensors exhibited a sensitivity of 1.2 pm/ $\mu \varepsilon $ . In bending tests of the glove, the linearity was 0.98961, the bending sensitivity was 1.8796 pm/°, and the maximum wavelength shift after bending reached 1.7 nm, demonstrating excellent stability and sensitivity. We also designed 16 common underwater communication gestures and six different complex usage scenarios, collecting datasets using the FBG-SenseGlove. A self-designed GRU-FusionNet deep learning network model was employed to extract, fuse, and classify the temporal features of the dataset. The results show that the validation set accuracy for gesture recognition in undisturbed normal scenarios was 99.48%, and the accuracy on the test sets in complex scenarios reached 98.98% after model transfer. Our FBG-SenseGlove, combined with deep learning algorithms, operates stably even in complex environments, providing theoretical validation and technical support for underwater gesture communication and specialized tasks in extreme conditions.