FBG Tactile Sensing System Based on GAF and CNN

Abstract

In recent years, tactile perception has attracted more and more attention as one of the important sensing technologies. Fiber Bragg grating (FBG) can be used as an advanced tactile sensing element based on the change of wavelength reflection spectrum under the tiny tactile force, which also has the characteristics of its small size and the fact that it is easy to be encapsulated in the industrial manipulator. This article proposes an object classification scheme for the FBG tactile sensing system based on the Gramian angular field (GAF) algorithm and convolutional neural network (CNN). Three identical FBGs are pasted on the surface of a flexible three-claw manipulator to obtain a three-channel tactile sensing signal, which is demodulated by the structure of wavelength-swept optical coherence tomography. In principle, any number of channels is applicable. The FBG tactile sensing signal belongs to 1-D small volume data, which transmits fast and occupies a small bandwidth. GAF maintains the correlations of time stamp during the process of encoding 1-D time series into 2-D images. CNN extracts deep features of data without a manual sign. Four typical CNN models are compared, which shows the feasibility of the proposed scheme. Finally, Resnet18 is chosen as the classifier of six types of object, and the accuracy of classification can reach 99.75% and the classification response time is only about 1.1 ms, which is suitable for application in any scenes, especially in smart industry with precious bandwidth, high accuracy, and low delay requirements.