Palm sEMG-based user identification during doorknob rotation using a convolutional neural network

Convenient and secure user identification is increasingly important in everyday environments, particularly with the proliferation of contactless interactions and Internet-of-Things (IoT) devices. However, conventional authentication methods often require explicit user input or additional hardware, limiting their usability in natural daily scenarios. To address this issue, we propose a doorknob-rotation-based user identification method using palm surface electromyography (sEMG). sEMG signals were acquired from the abductor pollicis brevis and abductor digiti minimi at 1,000 Hz, denoised using a 60 Hz notch and 20-500 Hz band-pass filters, and transformed into time-frequency spectrograms via continuous wavelet transform. A DenseNet161 model was employed for classification. Using data from five participants, the proposed method achieved 94.00% test accuracy and 93.99% F1-score, with five-fold cross-validation accuracy of 91.66[Formula: see text]2.78%. The approach enables on-device, contact-based identification without wireless pairing, transforming everyday actions into seamless authentication. These results demonstrate the feasibility and practical potential of sEMG-based everyday-action user identification.