Research On Object Classification Method Of Multi-tactile Sensor Sensing Data

The sense of touch is an irreplaceable source of information when humans explore the surrounding environment.It transmits various sensory information such as pressure,vibration and temperature to the central nervous system,and has advantages over vision and hearing in processing the material characteristics and detailed shapes of objects.Give the robot tactile perception ability to perceive the surrounding environment and provide useful information for subsequent tasks(such as classification of targets).Therefore,the study of object classification through tactile perception is of great significance to the leapfrog development of intelligent robots.Based on pure tactile data collected by multiple tactile sensors,this paper studies a variety of target classification methods.First of all,in view of the poor classification of multi-targets(more than 20 types of targets)of pure tactile perception data,this paper studies 41 types of target classification methods based on multiple classic machine learning algorithms.In this paper,the parameters of the kernel function of the support vector machine are improved,and the accuracy of multiobject classification is improved through the differential evolution algorithm.Through the comparison of the accuracy,precision,recall and F1 value of the four comprehensive evaluation indicators of machine learning algorithms,the advantages and disadvantages of each machine learning classification algorithm in the multi-object classification problem of tactile perception data are analyzed.Then,this paper verifies that a variety of machine learning algorithms are effective in the application of multi-tactile sensor sensing data.Then,for the tactile sensing capture data features of different sizes,different shapes,and different hardness multi-targets,the characteristics of the data are more complicated,the model training effect is unstable,the target classification is easy to be confused,and the classification accuracy is low.This article combines convolutional neural networks and deep residual network,an improved convolution residual network model(CN-Res N Model)is proposed and optimized.At the same time,the K-means cluster analysis method is used to process the tactile perception data,which strengthens the characteristics of tactile information.Through the fusion of the CN-Res N model and the K-means clustering analysis method,the classification accuracy of 26 types of targets with complex tactile characteristics is improved,and the application of the CN-Res N model in more complex tactile sensing data capture data is realized.A large number of experiments have proved the effectiveness and feasibility of the proposed target classification method.