Electronic skin pressure map mask generation based on optimized Real-ESRGAN-MSDC

Electronic skin (e-skin) visualizes contact pressure distribution through pressure maps, offering significant potential for robotic tactile sensing and health monitoring. However, insufficient sensor resolution often causes image shape distortion, limiting its use in high-precision applications. We propose an optimized Real-ESRGAN-MSDC model to generate high-fidelity shape masks from distorted pressure maps. Experiments are conducted using 432 image pairs with three rounds of 5-fold cross-validation. Performance differences between traditional computer vision mask generation (TCVMG), Pix2pix, CycleGAN, ESRGAN, Real-ESRGAN, and Real-ESRGAN-MSDC are evaluated using four metrics. The results demonstrate that Real-ESRGAN-MSDC outperforms other models, achieving 1.00% higher IoU, 0.59% higher F1 score, 0.70% higher SSIM, and 0.38% lower LPIPS compared to the second- ranked model, indicating improved structural similarity and edge fidelity. Our study provides novel insights into intelligent applications of e-skin for tactile sensing and health monitoring. Future work could expand the dataset, optimize training strategies, and reduce computational complexity to enable arbitrary shape prediction and real-time applications.