Texture Recognition of the MWCNTs/PDMS Flexible Tactile Sensor Based on Machine Learning Models

Texture recognition based on tactile sensors plays an important role in human-machine interaction. In this work, a novel flexible piezoresistive tactile sensor with high sensitivity (1.532 N⁻¹) and fast response (20 ms) ability is designed and fabricated based on multiwalled carbon nanotubes (MWCNTs)/polydimethylsiloxane (PDMS) nanocomposites. The sensor can precisely detect tactile information of different types of textures. The output voltage time series of the sensor from different textures is collected as the raw dataset. The discrete wavelet transform (DWT) and RF-recursive feature elimination (RF-RFE) methods are combined to select the optimal features for the textures based on the raw dataset. The RF model is constructed to recognize the different types of textures based on the optimal features. The recognition results of the RF model with the evaluation factors of accuracy, precision, recall, and F1 measure are very remarkable, which are 99.37%, 99.39%, 99.37%, and 99.38%, respectively. Identically, the support vector machine (SVM) model and the K-nearest neighbor (KNN) model are built based on the same datasets. The texture recognition results of the RF model are much better than the results of the SVM model and the KNN model. All the experimental results show that the sensor with great sensation capability can efficiently perceive and distinguish different texture signals, and the RF model with great ensemble learning ability is very suitable to recognize different texture signals collected from the sensor and gain excellent recognition results.