FEM‐Assisted Deep Learning for Object Recognition in Wearable Smart Gloves

Abstract Wearable smart gloves have transformative potential in fields such as robotics, rehabilitation, and human‐computer interaction, enabling precise gesture recognition and pressure sensing. However, sensor noise, environmental factors, and user‐specific differences hinder their real‐world application. This work presents an innovative framework that combines finite element modeling (FEM) with machine learning (ML) to improve pressure prediction and object recognition in smart gloves. FEM simulations accurately predicted contact pressure distributions during hand‐object interactions, with a mean absolute error (MAE) of 0.0184 MPa and high correlation with experimental data (Pearson's r = 0.875, Spearman's ρ = 0.971). Using this data, a backpropagation neural network (BP‐NN) was trained, significantly enhancing object classification accuracy across 21 categories. A baseline model (Model 1) performed well but struggled under noisy conditions. Model 2, incorporating noise reduction techniques such as Gaussian noise injection, grip force variability and sensor dropout, achieved a 95.05% classification accuracy, demonstrating improved adaptability to real‐world variability. This approach bridges FEM and ML to create a robust framework for smart gloves, addressing key limitations and enabling scalable, high‐accuracy solutions in robotics, virtual reality, and healthcare systems.