Data Analysis of Textile-Based Sensors Based on AI Algorithm
TL;DR
A hybrid AI framework integrating CNN– Transformer architectures to address challenges of textile-based sensors through adaptive wavelet denoising, GAN-based data augmentation, and attention-guided multimodal feature fusion for heterogeneous sensing signals is proposed.
Abstract
Textile-based sensors have emerged as key components for wearable healthcare, intelligent sports analytics, and industrial monitoring, yet their practical deployment remains constrained by electromagnetic interference, motion artifacts, nonlinear multimodal coupling, and limited computational efficiency in dynamic environments. This study proposes a hybrid AI framework integrating CNN– Transformer architectures to address these challenges through adaptive wavelet denoising, GAN-based data augmentation, and attention-guided multimodal feature fusion for heterogeneous sensing signals. A lightweight deployment strategy combining model pruning and quantization reduces computational overhead by 70%, enabling real-time inference with latency below 50 ms on edge devices. Experimental validation demonstrates that the proposed framework achieves 95.3% motion recognition accuracy using an 8 × 8 pressure sensor array, attains an AUC of 0.95 for respiratory anomaly detection under low signal-to-noise conditions, and predicts fabric remaining useful life with an RMSE of 6.2 hours through physics-informed learning. By improving robustness against environmental disturbances and enhancing multimodal information extraction, the proposed method provides an effective computational solution for intelligent textile sensing systems and offers methodological support for electromagnetic interference mitigation, wearable sensing networks, and advanced smart sensing platforms involving wireless signal acquisition and integrated electromagnetic environments.