Research On Spatio-temporal Correlation Feature Extraction And Recognition Of Multi-modal Tactile Signals

With the rapid development and maturity of artificial intelligence technology,recognition of multi-modal tactile signals by robots has become a hot topic in the field of autonomous robotics.Multi-modal tactile signal recognition is one of the most important ways for robot to interact with the external environment.It is the key of robot to understand tactile signal effectively.It is the basis for intelligent robot to accurately judge external environment information and make strategy selection and behavior control.At present,the recognition accuracy of multi-modal tactile signals is to be improved,the main reason is that the feature extraction method has a shortage of spatio-temporal correlation features in tactile signal mining.The classification method for tactile signal recognition should be improved in the multi classification problem.In this paper,the multi-modal tactile signal acquisition system is built to collect the experimental data.By mining the temporal and spatial correlation features in the signal,the binary tree structure support vector machine(SVM)is optimized,and the accuracy of multi-modal tactile signal recognition is improved.The main research work is as follows:(1)Design and implementation of multi-modal tactile signal acquisition system.A multi-modal haptic signal is acquired using an FSR sensor array in a three-dimensional space,and the force direction of the experimenter is corrected in real time by a loadcell,and an experimental platform capable of sensing multi-modal signals is constructed.The problem of asynchronous time synchronization between FSR sensor and loadcell sensor is solved on the upper computer,and the acquisition and preservation of the tactile signal are completed,and the control interface of data acquisition is realized.It lays an experimental foundation for further theoretical research.(2)A spatio-temporal correlation feature extraction method for multi-modal haptic signals is proposed.Before the feature extraction,the adaptive Calman filter is used to filter the data and reduce the interference of noise to tactile signal recognition.The extraction of spatio-temporal correlation features first uses CNN to extract spatial correlation features of different tactile sensor signals,and simultaneously cuts the continuous dynamic tactile signals into a number of short tactile signals,and then extracts spatio-temporal correlation on the basis of the spatial characteristics of the multiple time fragments using the GRU's correlation characteristics on the time series.The experiment shows that the spatio-temporal correlation feature extracted by the methods can get higher classification accuracy in various classification algorithms compared with the features extracted by other methods.(3)A method of optimizing binary tree structure support vector machine is proposed to improve the accuracy of multi-modal tactile signal recognition.The binary tree structure is optimized by the improved particle swarm optimization(PSO)algorithm,and the error accumulation of the multi-classifier of the binary tree structure is reduced.Experiments compared the accuracy of the proposed method and other methods for multi-modal tactile signal recognition,and the results proved the effectiveness of the proposed method.