Frequency-response Analysis Of The Wheat Acoustic Test Signal

The Merits of Wheat quality directly affect its use value and economic value. Quality inspection plays very important role in circulation links of wheat such as purchase, processing, storage and sales. In view of the existing wheat quality detection technology can not meet the requirements of classification quickly, the establishment of an objective, fast and accurate wheat quality detection method is of great significance. In this paper through analysis the relationship between the acoustic characteristics of the wheat kernels and hardness index, preliminary study the frequency characteristics of wheat acoustic signal, through the analysis of the difference between different frequency bands of wheat sound signal, optimize the wheat quality acoustic detection method.On the basis of previous studies, built the experimental platform of wheat sound signal acquisition device, collected the wheat sound signal of different frequency band, and through the analysis and processing of the wheat sound signal in time domain and frequency domain, extracted some characteristic parameters which have good correlation with hardness index. In time domain, extracted some characteristic parameters such as waveform index TF1, pulse factor TF2, peak factor TF3, and the frequency band which have best correlation between the feature parameters and hardness index were respectively40～60kHz,25～40kHz and25～40kHz. In frequency domain, extracted some characteristic parameters such as FER based on fast Fourier transform, DCT2based on discrete cosine transform, WF based on the wavelet transform, WPT based on wavelet packet transform, and the frequency band which have best correlation between the feature parameters and hardness index were respectively25～40kHz,30～50kHz,35～70kHz and25～40kHz. Choose some suitable characteristic parameters, through linear regression analysis and neural network technology, built the corresponding wheat hardness acoustic detection model. Through the comparison and analysis of the test results, discussed the difference of detection feature between the different frequency bands. In25～40kHz,35～70kHz,30～50kHz,40～60kHz and50～70kHz frequency band, the average relative error of the optimal linear regression model detection results are respectively5.99%, 8.96%,7.82%,8.00%,10.26%; the average relative error of Neural network models to detection results are respectively1.96%,3.08%,7.07%,3.61%,7.03%. This suggests that in the same frequency band, the detection effect of neural network models is superior to linear regression model; the same prediction model at different frequency bands is difference, the detection precision of neural network model of25～40kHz is superior to the other frequency bands’detection model.