A Deep Learning-Based EIT System for Robust Gesture Recognition Under Confounding Factors

Gesture recognition with electrical impedance tomography (EIT) is an enormous potential tool for human–machine interaction because of its low cost, low complexity and high temporal resolution. Although high-precision EIT-based gesture recognition has been achieved in ideal scenarios, ensuring its consistent performance under interference remains challenging. This article presents a novel method to alleviate the effect of confounding factors on EIT gesture recognition. An EIT armband was designed to mitigate the effect of contact impedance variation based on equivalent circuit analysis, and a spatial–temporal fusion network, named the Fold Atrous Spatial Pyramid Pooling-Gated Recurrent Unit (FASPP-GRU), was developed for robust gesture classification. The results showed that the proposed two-layer electrode maintained a stable contact impedance when its contact force with the skin was changed. Although confounding factors caused significant changes in baseline forearm impedance, FASPP-GRU achieved 80% accuracy under the effect of limb position changes and dynamic changes in muscle state over time, which outperforms conventional classifiers. With an 87 μs inference time, the proposed system shows enormous potential in real-time applications.