Texture classification model based on temporal changes in vibration using wavelet transform

Tactile sensation is important in the perception of the external world, and research on texture classification using vibration data obtained by tracing an object has been widely conducted. However, few studies have utilized time-varying frequency components, which are thought to be recognized by moving their fingers back and forth when they feel tactile sensations. Therefore, we propose a new texture classification system that uses the time variation of vibration with which a latent vector effective in perceiving tactile sensations is possibly embedded. Vibration data was acquired by reciprocating the developed sensor with strain gauges and PVDF film on fifteen different samples. The wavelet transform of the vibration data was conducted to extract a scalogram containing time-varying information. A CNN was constructed to perform texture classification based on the scalograms, resulting in an accurate classification. The results also showed the robustness of the model regarding the vibration information against the different touch condition.