A flexible capacitive sensor with microstructure for handwriting recognition

Abstract

Abstract With the rapid advancement of information technology, human-computer interaction is undergoing transformative changes of unprecedented scale. Among various interaction technologies, handwriting recognition, being a natural and intuitive input method and consistently holds a significant position. A flexible capacitive pressure sensor, with the sensitivity of 5.734% kPa −1 , the response time of 200 ms, and a remarkable cyclic stability (>1000 cycles), is developed by the polydimethylsiloxane (PDMS) with microstructure. Through writing 26 letters ‘A ∼ Z’ on the sensor’s surface, 2600 sets of 200-dimensional capacitive time series signals are generated and collected, which are used to form the customized dataset for handwritten letters. Based on the dataset, a fusion model of Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM), CNN-LSTM, is constructed to improve the accuracy of handwriting recognition to 98.59%. To verify the efficiency the CNN-LSTM model, Random Forest (RF) and Support Vector Machine (SVM) are constructed to identify different handwritten letters based on the same dataset used for the CNN-LSTM model, and the corresponding recognition accuracies are 94.61% and 95.64%, respectively. All the experimental results demonstrate that the flexible capacitive sensor with great sensation ability can precisely detect and capture different handwritten signals, and the CNN-LSTM model with great feature extraction capability is very suitable to recognize different handwritten signals collected from the sensor.