随着畜禽肉和肉制品食用量的迅速增长, 人们对肉品质量提出了更高的要求; 对于肉制品, 消费者最为关心是肉品质量, 当前中国对肉品品质在线检测方面的研究和应用则相对较少, 尚无针对肉品品质在线无损检测开发的设备。 也没能真正投入到肉品的生产加工过程。 研究不同肉品脂肪的近红外快速检测模型。 并采用标准化学方法进行差异分析。 通过近红外技术对猪肉、 牛肉、 羊肉进行扫描, 采用国标法（索氏提取法）对鲜肉脂肪含量进行化学值的测定, 以PLS（偏最小二乘法）作为建模方法, 并通过不同的光谱预处理手段分别建立了猪牛羊肉的近红外光谱参数与样品的脂肪含量之间的对应关系模型。 结果表明, 对于猪肉来说, 选择4 260～6 014 cm-1波段+一阶导+Norris所建的模型效果最好, 其校正相关系数和预测相关系数分别为0.955 6和0.961 6; 对于牛肉来说, 选择5 226～7 343 cm-1波段+一阶导+S-G所建的模型效果最好, 其校正相关系数和预测相关系数分别为0.923 5和0.942 7; 对于羊肉来说, 选择5 207～7 362 cm-1波段+一阶导+Norris所建的模型效果最好, 其校正相关系数和预测相关系数分别为0.915 7和0.939 6; 对于鲜肉来说, 选选用波段为5 156～6 065 cm-1+二阶导+S-G所建模型效果最好, 其校正相关系数和预测相关系数分别为0.916 3和0.919 4。 以上所有模型的校正相关系数均大于0.91, 模型都具有较高的精密度, 符合不同肉制品在实际生产的需求, 具有分析速度快、 检测成本低、 分辨率高、 无损的优点。

Meat and meat products are important components of the human food chain. Their quality is an important issue to consumers, government agencies and retailers. In China, the research and application of rapid and reliable methods for on-line detection of meat quality is still in badly need. Near infrared spectroscopy (NIRS) is an attractive technique for such applications, since it is fast, non-destructive method which requires small samples with a high-penetration radiation beam and free from further preparation of the samples is needed. Therefore, in this study, the overall objective was to investigate the use of a NIR hyperspectral imaging technique for accurate, fast and objective detection of fat content in various meat, and to compare with traditional standard chemical results. With near infrared scanning technology to pork, beef, mutton, and the national standard method (soxhlet extraction method) to determination of chemical values of fresh meat fat, with PLS (partial least squares) as a modeling method, and through the different spectral preprocessing methods respectively established the pigs, beef and mutton samples of near infrared spectrum parameters and the corresponding relationship between the fat content of model. Results show that for pork, select band 4 260～6 014 cm-1+a derivative+Norris derivative model built by the best effect, the correction coefficient of correlation and prediction correlation coefficient of 0.955 6 and 0.961 6 respectively; For beef, choose 5 226～7 343 cm-1 band+a derivative+S-G model built by the best effect, the correction coefficient of correlation and prediction correlation coefficient of 0.923 5 and 0.942 7 respectively; For mutton, select band 5 207～7 362 cm-1+a derivative + Norris derivative model built by the best effect, the correction coefficient of correlation and prediction correlation coefficient of 0.915 7 and 0.939 6 respectively; For fresh meat, choose select band of 5 156～6 065 cm-1+second derivative+S-G model built by the best effect, the correction coefficient of correlation and prediction correlation coefficient were 0.916 3 and 0.919 4, above all model correction of the correlation coefficient is greater than 0.91. Thus, the model has higher precision, meeting the needs of different meat products in the actual production.It is a nondestructive method with advantages such as fast analysis speed, low cost and high resolution.