Study On The Acoustical Determination Method Of Wheat Hardness

Wheat hardness is an important indicator of evaluating the qualities of wheat.This paper is to study the relationship between acoustic features of wheat kernels andwheat hardness. Analyze sound signal from wheat hit by using signal processingtechnology. From the signal, relevant feature parameters about wheat hardness can begot to construct the acoustic determination model of wheat hardness, which can laytheoretical basis for developing the acoustic determination instrument of wheathardness.Study the acoustic features of wheat kernels, and design the device to collect thesound signals of wheat. Made the wheat automatic feeder, so that the wheat fallnaturally and hit the target one by one. Receive the sound signal from wheat hit byusing the microphone, and modulate the sound signals of wheat using of the signalconditioner and amplifier modulation. Perform Analog-Digital Conversion of thesound signal of wheat by using the acquisition card, and pretreatment them. Analyzethe features of sound signals of wheat in the time domain and frequency domain byusing signal processing techniques, and get feature parameters which correlated wellwith hardness index. In the time domain, extract feature parameters from the soundsignal of wheat such as zero rate (TF5) and wave indicator (TF6) and pulsefactor(TF7) and so on. The correlation coefficients of TF5, TF6, TF7with wheathardness are respectively0.80,0.92, and0.89. In the frequency domain, from thesound signal of wheat, extract features FERa, FERband FERc, which are based on thefast Fourier transform. The correlation coefficients of FERa, FERb, FERcwith wheathardness are respectively-0.89,-0.88,-0.92. Extract the feature of DF1from thesound signal of wheat, which is based on the Discrete Cosine Transform. Thecoefficient of correlation between DF1and wheat hardness is0.83. Extract WF1andWF2, which are based on the Wavelet transform. The correlation coefficients of WF1and WF2with wheat hardness are respectively-0.93,-0.92. Research on therelationship between the extracted features and the wheat hardness by using of linearregression and neural network technologies, establish the acoustic prediction model of wheat hardness, and analyze the predicted results of the models. Finally, make the BPneural network prediction model which is based on the feature of WF1as the acousticdetermination model of wheat hardness. The maximum relative error predicted byresults of this model is-2.24%and the average relative error is0.15%. The predictionresults of the models show that, it is feasible for determining wheat hardness by usingthe acoustic method.