Smart Insole Pressure Distribution Prediction and Gait Optimization System Based on Multimodal Data Fusion and ViLT

To improve the accuracy and adaptability of gait analysis in embedded smart insole systems, this study proposes a multimodal pressure distribution prediction and gait optimization framework based on Vision-and-Language Transformer (ViLT) and cross-modal spatiotemporal modeling. The proposed approach addresses the limitations of modality isolation, temporal misalignment, and rigid fusion strategies by integrating plantar pressure sensing, inertial measurement unit (IMU) signals, and biomechanical parameters into a unified end-to-end architecture. A dynamic attention alignment mechanism is developed to establish semantic associations among heterogeneous data streams, while an LSTM-based spatiotemporal coupling model combined with hyperelastic mechanics enables accurate pressure prediction and personalized gait optimization. Considering that wearable sensing systems increasingly rely on electromagnetic signal acquisition and wireless data transmission, the proposed multimodal fusion framework also provides methodological support for intelligent sensing architectures employing antenna-enabled communication and electromagnetic information acquisition. Experimental results demonstrate that the proposed system achieves regional pressure prediction errors below 0.9% for diabetic foot subjects and approximately 0.8% for healthy subjects, while reducing peak plantar pressure by more than 12% and 8%, respectively. The hardware–algorithm co-design and closed-loop feedback strategy significantly enhance gait adaptability and rehabilitation effectiveness, providing a reliable computational framework for intelligent wearable sensing, multimodal electromagnetic data fusion, and nextgeneration smart healthcare applications.