Surface texture recognition network based on flexible electronic skin

Abstract

The material of an object can be identified through the analysis of its surface texture. However, it is not easy to determine uneven surface texture via machine vision. The information collected by tactile sensing can well retain the concave and convex features of texture itself, which makes this type of texture recognition relatively simple. In this paper, we design three definitive artificial texture models, and use capacitive flexible tactile perception electronic skin to collect pressure data. We apply constant mass loads on the electronic skin, fully touching the surface to be detected, and then drag the electronic skin to produce relative sliding horizontally which to ensure the stability of the collected data in the vertical direction. Based on the convolutional neural network a classification neural network is constructed. The tactile pressure information is input of the neural network after special processing so that it still retains the spatial and temporal information at the same time. We have completed the detection of the deterministic surface texture of certain objects, and the final accuracy can be achieved more than 90%, which shows that the network has a good recognition ability for the surface texture.