Research On Object Surface Material Recognition Based On Auditory Feature Analysis

With the rapid development of artificial intelligence,robots are widely used in many fields such as home services,chemical engineering,and aerospace.The target object material recognition is an important part of the robot's perception of the external environment and target object recognition.As the robot's two apperceive modalities,vision and hearing play a decisive role in the robot's perception of the environment.However,when the detected object is visually similar and the constituent materials are different,the analysis and recognition results of the robot vision system may be incorrect,and the auditory modes may compensate for the lack of vision in this respect.In this regard,this study applies the voice recognition technology to the surface material recognition of the object,analyzes and recognizes the surface material of the object by analyzing the sound generated by striking the object or rubbing the surface of the object,so as to achieve the purpose of environmental perception.In this paper,we designed a microphone sound acquisition pen for material recognition of the surface of the object.Using the sound collection pen,five kinds of material percussion sound data sets TSD-25 and nine wood percussion and sliding sound data sets WSD-9 are set up respectively,and corresponding experimental analyses were conducted respectively:(1)In this paper,three kinds of sound features including high-order Mel cepstrum coefficients(HMFCC),higher-order linear prediction Mel cepstrum coefficients(LPHMFCC),and high-order Mel cepstrum coefficients based on wavelet transform(WHMFCC)are extracted from the TSD-25 data set.Then,the relative importance of the cepstrum components of each of the obtained sound features is evaluated using the method of increasing and decreasing components to achieve the feature screening effect.At last,two kinds of machine learning algorithms,support vector machine and extreme learning machine,are used to analyze and compare the surface texture recognition results of the three kinds of sound features.Experiments show that the LPHMFCC feature has the best recognition effect among the three kinds of sound features,and the extreme learning machine has a better recognition effect than the support vector machine.(2)For the WSD-9 datasets,the three features and features of the HMFCC,LPHMFCC and WHMFCC are extracted,and the features of the screening are combined with multiple features.Finally,the processed sound feature data is studied based on sound multi-feature combination and RELM algorithm.The experimental results show that the recognition effect of multi-feature combination of sound is more advantageous than the recognition effect of single-sound feature,and the single sound mode based on the combination of multi-sound mode is more advantageous.However,the sliding sound is slightly inferior to the percussion sound in the wood material recognition.In this study,we also designed a surface material recognition system through the GUI of MATLAB software to make it more simple and convenient in the application process.