Optimal Prefiltered Stochastic Transmitted Waves for Fat Inclusion Detection in Milk with Harmonic Ultrasound

In non destructive testing, improvements have been made possible by taking into account the harmonic frequencies, as in agri-food domain. The transmitted signal are often selected empirically as a fixed-frequency Gaussian pulse, by taking into account the transducer bandwidth only. However, waveform should take into account all the features of the ultrasound system and of the medium. To design the waveform, a genetic algorithm looks for the best stochastic wave. However, one of limitations in this optimization process is the high amount of transmitted waves. To reduce this number, instead of transmitting wide band stochastic waves, narrow band stochastic waves limited by the transducer bandwidth are preferred. The optimization was thus applied on the detection of fat cluster in milk by maximizing the signal-to-noise ratio (SNR), while decreasing the amount of transmitted waves. Twelve combinations from different limited bandwidths of transmitted waves were tested. Whereas the low cut-off frequencies did not change the performances, the high cut-off frequencies affected the convergence speed. In this study, it is shown that the best optimization was twelve times faster with the high cut-off frequency of 5.6 MHz and led to a gain of 62% compared to the SNR obtained with a best fixed-frequency sine wave.