基于近红外光谱的散养土鸡与笼养肉鸡的营养成分预测

Nutrients components forecast on free-range and broiler chicken based on near infrared spectroscopy

目的建立土鸡与肉鸡的蛋白质、脂肪含量快速预测模型。方法收集土鸡与肉鸡新鲜样本各30份,取其中各20份样品,应用近红外光谱分析技术和区间最小二乘法建立蛋白质、脂肪的定量分析模型;然后对剩余样品进行预测,并进行误差分析。结果土鸡与肉鸡的蛋白质模型相关系数分别是0.978和0.963,内部交叉验证均方差(RMSECV)分别为0.197和0.201;脂肪模型的相关系数分别为0.946和0.952,RMSECV值分别为0.318和0.149。预测中,蛋白质预测结果与实测结果误差平均为0.193和0.214,标准差为0.098和0.065;脂肪预测结果与实测结果的误差平均值分别为0.318和0.149,标准差分别为0.072和0.103。结论通过预测结果与实测结果比较,发现差异并不显著,标准方差在10%及以下,并且预测模型的相关系数越大,预测结果越准确,说明了近红外光谱技术与区间最小二乘法预测模型的可行性、准确性、快速便捷性,能够为市场土鸡肉与肉鸡肉的鉴别提供快捷有效的方法。同时,为提高预测结果的准确性,需采用尽量多的样品建立预测模型。

Objective To build a fast predictive model about protein and fat content of chicken and broilers. Methods Thirty fresh samples from both chicken and broilers were collected separately and 2 quantitative analysis models were built for determination of protein and fat content by using 20 samples of each. The remaining 20 samples were analyzed by predictive analytics and error analytics. Results The correlation coefficients of protein model with 2 kinds of chicken were 0.978 and 0.963 when RMSECV were 0.197 and 0.201 in the chicken and broilers protein model, respectively. As for the fat model, the correlation coefficients of 2 kinds of chicken were 0.946 and 0.952, and RMSECV were 0.318 and 0.149, respectively. It turned out to be that the mean errors of predicted and actual outcomes were 0.193 and 0.214, the standard deviations were 0.098 and 0.065 in protein’s case, respectively. And mean errors of predicated and actual outcomes were 0.318 and 0.149, the standard deviations were 0.072 and 0.103 in fat’s case, respectively. Conclusion From the compared results, the difference between predicted and actual results was not significant, and the standard deviation was 10%or lower. In addition, the larger the correlation coefficient of the prediction model, the more accurate the prediction results. At the same time, it showed that fast predictive models based on NIR analysis technique and the least square method can provide efficient feasible and accurate approaches to the identification of chicken and broilers. Then, it is necessary to collect more samples to built the prediction model for improve the accuracy of forecasting results.