Classification Of Different Materials Using Their Acoustic Signals

Abstract

An acoustic signal transmitted through a material gives a lot of qualitative information about itself. Such information can be useful in classifying different materials. The present paper discussed a new method to classify liquid mixtures without having knowledge of solutes. For this purpose acoustic signals of different liquid mixtures were acquired using a simple acoustic resonance spectrometry system. The spectrometry system has been developed by a V-shaped quartz tube and two similar piezoelectric transducers. The transducers are attached at both the ends of the tube. The white noise signals were given to the transmitter transducer which generates vibrations. The generated vibrations transmitted through the quartz tube and the liquid mixture sample. The vibrations after interaction with the sample were translated into an equivalent voltage signals by the detector transducer. These signals have been recorded and analyzed by Laptop and software. Three types of liquid samples, namely: water, salt solution, and sugar solution were used in the experiments. The spectral features of each material were extracted from recorded signals by autoregressive power spectral density. These features were given as inputs to the classifiers: SVM, QDA, and KNN. The overall classification accuracies of QDA, KNN, and SVM were found to be 98%, 99.6%, and 100% respectively when all the spectral features had been given to classifiers. The results show that the SVM classifiers provide the best classification accuracy on the autoregressive spectral features of materials.