Tactile Perception Object Recognition Based on an Improved Support Vector Machine

Abstract

Tactile perception is an irreplaceable source of information for humans to explore the surrounding environment and has advantages over sight and hearing in processing the material properties and detailed shapes of objects. However, with the increasing uncertainty and complexity of tactile perception features, it is often difficult to collect highly available pure tactile datasets for research in the field of tactile perception. Here, we have proposed a method for object recognition on a purely tactile dataset and provide the original tactile dataset. First, we improved the differential evolution (DE) algorithm and then used the DE algorithm to optimize the important parameter of the Gaussian kernel function of the support vector machine (SVM) to improve the accuracy of pure tactile target recognition. The experimental comparison results show that our method has a better target recognition effect than the classical machine learning algorithm. We hope to further improve the generalizability of this method and provide an important reference for research in the field of tactile perception and recognition.