Abstract:Biospeckle is one of the low-cost, portable and online screening tools for optical non-destructive testing technologies, and it shows potential for application to agricultural products quality prediction. Sensory evaluation, texture profile analysis (TPA) and Warner—Bratzler (W—B) shear force were applied to analyze the texture characteristics of beef tenderloin, the correlation between different measuring methods was investigated, and the prediction model of biospeckle for texture characteristics was established. Since the significant difference between tenderloin and shin, it seems not possible to predict their texture characteristics with a same model. Two methods, including slope/bias (S/B) correction method and Kennard—Stone (K—S) typical samples adding method were used to improve the tenderloin prediction model. Compared with the effect of two modified methods, the more accurate and convenient method was chosen to make the model transfer to shin fast. The results showed that the hardness and chewiness of sensory evaluation and TPA had high positive correlation, the determination coefficient (R2) reached 0.98 and 0.90, respectively, and R2 between W—B shear force and hardness of TPA reached 0.95, which proved the reliability of the three texture characteristics measurement methods. The values of R2 for predicting the texture characteristics of hardness, chewiness and W—B shear force with biospeckle activity were 0.83, 0.77 and 0.69, respectively. The results of improvement for the loin model were as follows: as improved with S/B correction method, the root mean square error (RMSE) was 26.65, bias factor (Bf) and accuracy factor (Af) were 1.08 and 1.15, respectively. While the effect of modified with K—S adding method of typical samples was better than that of S/B correction method, and when the adding number of samples was 12, the RMSE was 13.21, Bf and Af values were 1.07 and 1.02, respectively. In conclusion, K—S typical samples adding method could reduce the differences between the different parts, improve the goodness-of-fit of predictive shin model, and produce better effect than S/B correction method.
