本文在近红外反射光谱780~1700 nm的波长范围内采集新宰杀的同一品种的羊的后腿肉134个样本的光谱数据，来实现快速无损的南疆生鲜羊肉含水量的检测。这些光谱数据经中值平滑滤波、多元散射校正、一阶导数、标准化处理、中心化变换和S-G平滑等预处理方法对原始光谱进行降噪处理；然后以13:1的比例将样本分为训练集和测试集，并采用PLSR建立预测模型，使用所建模型对生鲜羊肉水分含量进行预测。结果为：训练集的预测相关系数Rc为0.94、标准差MSEC为0.04，预测成功率为97.6%，测试集的预测相关系数Rv为0.89、标准差MSEV为0.07，预测成功率为96.4%。实验结果证实结合中值平滑滤波、多元散射校正、一阶导数、标准化处理、中心化变换和S-G平滑等多种预处理方法建立的基于近红外光谱PLSR模型，可以对南疆鲜羊肉的水分含量进行精确的快速无损评价，并且能为南疆生鲜羊肉水分含量的快速无损检测技术的应用提供理论上的指导。

In order to realize fast nondestructive testing of moisture content fresh mutton in southern Xinjiang, the spectroscopic data of the 134 samples of same variety in fresh hind legs meat of sheep in the range of 780~1700 nm wavelength were collected in this study. The original spectrums were processed by noise reduction such as median smoothing filtering, multiple scatter correction, a derivative, standardization, centralized transformation, smoothing retreatment methods, and S-G smoothing. The samples were divided into training set and test set by a ratio of 13:1. PLSR was adopted to establish the prediction model to forecast the moisture content of the fresh mutton. The results showed that the prediction correlation coefficient of training set was 0.94, its MSEC was 0.04, and prediction success rate was 97.6%, while the prediction correlation coefficient of the test set was 0.89, its standard deviation MSEV was 0.89, and the success rate of prediction was 96.4%. Experimental results confirmed that the established PLSR model combined with preprocessing such as median smoothing filtering, multiple scatter correction, a derivative, standardization, centralized transformation, and the S-G smooth could provide accurately fast nondestructive evaluation for the moisture content of fresh mutton, and provide theoretical reference application for rapid nondestructive testing of the moisture content in fresh mutton.

国家青年自然基金资助项目（61640413）