Self-powered flexible handwriting input panel with 1D output enabled by convolutional neural network

The growing needs for wearable electronics urge the development of smart human-machine interfaces. Multi-output channels are required for current flexible input panels to realize trajectory detection and user identification functions. Herein, a self-powered flexible input panel with 1D output for multifunctional input detection, including letter recognition, user identification, and digit pattern detection, is proposed. The input panel is ideal for wearable human-machine interface owing to the good conformability of PU membrane to human skin and the robust performance under bending state. A 1D convolutional neural network is designed and optimized to achieve a classification accuracy of 97% on 7 letters and identification accuracy of 96.3% on five participants based on the triboelectric output from the spiral carbon grease electrodes pair of the proposed device. Demonstrations of harvesting energy from fabric contact and real-time digit pattern recognition are proposed to show the potential applications of the proposed input panel. These results generate fresh insight into wearable smart input panel design. A self-powered flexible input panel with 1D output for multifunctional input detection, including letter recognition, user identification, and digit pattern detection, is proposed. The input panel is ideal for wearable human-machine interface owing to the good conformability of PU membrane to human skin and the robust performance under bending state. A 1D convolutional neural network is designed and optimized to achieve a classification accuracy of 97% on 7 letters and identification accuracy of 96.3% on five participants based on the triboelectric output from a pair of helix carbon grease electrodes. • A helix electrode design is proposed to reduce the number of output channels to 1. • Applying the sliding window segmentation method in the data preprocessing of a 1D convolutional neural network. • Chirography differences are extracted by a convolutional neural network from the sequential features in the triboelectric output. • A demonstration of real-time detection of handwritten numbers provides a reference for the design of wearable self-powered input detection devices.