Detection of wheat hardness based on a laser-generated ultrasonic signal

Hardness is an important index for wheat quality, which determines its usage, price and wheat-processing techniques. Therefore accurate measurements of the hardness of kernels is a key problem for assessing wheat quality. In this work we develop a new method for testing the hardness basing on a laser-induced ultrasonic signal. We describe the measurement principles, as well as sampling and preprocessing of ultrasonic signals. The acoustic signal is analyzed both in the time and frequency domains, using fast Fourier, discrete cosine (DCT) and wavelet (WT) transforms. The main eight parameters whose correlation indices surpass the threshold of 0.8 are selected as feature characteristic parameters of the hardness. They include a waveform index T6, a pulse factor T7, sub-band energy ratios SER1-SER3, a sum of the DCT magnitudes, and two wavelet parameters, WTF1 and WTF2. A testing model of the hardness is built basing on a standard extreme learning machine (ELM) algorithm and using all of these feature parameters as an input. A number of experiments have been performed on twenty wheat varieties with different hardness indices. The results show that the maximal relative measurement error and the mean relative error are approximately equal to -3% and 1%, respectively. As a result, our method of measuring the wheat hardness, which is based on combination of laser ultrasonic waves and ELM, is feasible and accurate enough.