| As the development of husbandry and feed industry, the quality and safety of feed products have become more and more important. The evaluation of feed quality involves two aspects, one is nutrition, and the other is processing. In the circumstance of the feed ingredients formulation are fixed, processing quality not only influences the preservation and transportation, but also has effect on the utilization of nutrient, and then exerts negative effect on the production performance of animals. Pellet feed is a kind of important feed, so the quality evaluation is most of importance.This dissertation focused on bringing the Near Infrared Reflectance Spectroscopy (NIRS) to feed processing qualities evaluation, and got the conclusion that NIRS could be used to determine the quality of the semi-manufactured and finished feed. In the research, a total of 87 samples of different particle size were prepared to scan the near infrared spectra and build the quantitative model. Also, we blended feed and sampled at different time, and got the qualitative discrimination model of blending uniformity. Meanwhile, we collected 122 duck pellet feed samples and 106 porket pellet feed samples with different process techniques and feed formulation, made a detail research on the relationship of moisture, hardness and percentage of powdered pellets, and tried to build the prediction models of the above indices. As to the porket pellet feed, we determined the starch cooking degree and pasting viscosity values, and then got the quantitative model of cooking degree, and built the quantitative and qualitative relationship between starch cooking degree and pasting viscosity values. Specific results are as follows.(1) With Partial Least Square 1 (PLS-1) as the regression method and according to the special character of particle size, we didn't take any spectra preprocess methods, the calibration model was built. And then, we used the model to predict the validation set samples, in order to check out the prediction ability of the calibration model. The coefficient of determination (R2) of calibration set and validation set were 0.9463 and 0.9168, respectively. The Relative Standard Deviation (RSD) values were 6.56% and 4.38% (all less than 10%), respectively. The Relative Predictive Determination (RPD) values were 4.38 and 4.09(all over 3). The results showed that NIRS analysis technique could be adopted to correctly measure the corn particle size.(2) Chose discrimination as regression method, we built the qualitative model of blending uniformity of corn, soybean and 4% premix feed at the NIR range of 950-1650nm. The results showed that the seven principal components included over 99.9% of all sample information.The model was used to predict 29 unknown samples, the accuracy rate was 100%.(3) We built the calibration models with the duck pellet feed and the ground feed respectively, and made a comparison of the two models.The results showed that the percentage of powdered pellets couldn't be predicted accurately, for the R2 of calibration was very low. As to moisture and hardness, different model had different predictive ability. For the model from pellet feed, the R2 of calibration set of moisture and hardness were 0.9810 and 0.8977 respectively. The RSD were 1.71% and 8.26% (all less than 10%), respectively.The RPD were 7.25 and 3.01. The R2 of validation set of moisture and hardness were 0.9746 and 0.8179 respectively. The RSD were 1.71% and 2.95% (all less than 10%), respectively. The RPD were 6.30 and 2.56. For the model from ground feed, the R2 of calibration set of moisture and hardness were 0.9825 and 0.8799 respectively. The RSD were 1.39% and 9.05% (all less than 10%), respectively. The RPD were 8.69 and 2.89. The R2 of validation set of moisture and hardness were 0.9920 and 0.8024 respectively. The RSD were 4.41% and 11.05%, respectively. The RPD were 10.36 and 2.43. The statistic data indicated that the model from pellet feed had good prediction ability for both moisture and hardness, and the model from ground feed was weak in the prediction of hardness.(4) There are quantitative and qualitative relationship between starch cooking degree and pasting viscosity values. We found out the qualitative relationship and built the regression equation with the R2 value 0.6519.The calibration model of starch cooking degree was robust. The R2 of calibration set and validation set were 0.8759 and 0.9351, respectively. The RSD were 6.072% and 4.50% (all less than 10%), respectively. The RPD were 3.14 and 4.47(all over 3). The results showed that NIRS model was adopted to measure unknown samples.
|
PDF Full Text Request