利用近红外高光谱成像技术对滩羊肉蛋白质和脂肪含量、pH值进行无损检测研究。通过高光谱系统（900~1700 nm）采集69个羊肉样本信息，先对全波段下的原始光谱和预处理后光谱建立偏最小二乘回归（PLSR）模型，对比优选出最佳预处理算法，后采用PLSR的加权β系数法提取特征波长，建立特征波长下各品质参数的PLSR模型，分析预测效果。结果表明：羊肉蛋白质、脂肪含量、pH值最佳预处理方法为基线校准（Baseline）、多元散射校正与S-G卷积平滑结合算法（MSC+SG）和原始光谱；利用特征波长建立预测模型，决定系数（RP2）分别为0.83、0.86和0.72，预测均方根误差（RMSEP）为0.57、0.09和0.12，可替代全波段建模。利用近红外高光谱成像技术对羊肉内部品质进行快速无损检测是可行的。

In this paper, protein and fat content, pH value of Tan-sheep meat was nondestructively detected using near infrared hyperspectral imaging technology. Spectral information of 69 samples was collected by hyperspectral image system (900~1700 nm). The partial least squares regression (PLSR) models under full-wave spectrum established by the original spectrum were compared with pretreatment?ones, and the best pretreatment algorithms were selected. In addition, the characteristic wavelengths were selected through β weight coefficient?of PLSR, then the PLSR models of protein and fat content, pH value under the characteristic wavelengths were set up, and the prediction effects of models were analyzed. The results showed that: the best pretreatment algorithms for models of mutton protein and fat content , pH value were Baseline, MSC+ SG and the original spectrum; the determination coefficient (RP2) of models built under characteristic wavelengths were 0.83, 0.86 and 0.72, and the predict root mean square error (RMSEP) were 0.57, 0.09 and 0.12, which could replace the full-wave modeling. Thus, it is feasible for testing internal qualities of mutton quickly and nondestructively using NIR hyperspectral imaging technology.

国家自然科学基金资助项目(31101306)；“十二五”国家科技支撑计划课题(2012BAK17B07)