基于高光谱技术的羊肉含水率无损检测

利用高光谱成像系统(1000~2500 nm)对羊肉含水率进行无损检测研究。对108个羊肉样本进行光谱信息采集,通过标准正态变换法、归一化法、去趋势校正法、S-G卷积平滑法、导数法、多元散射校正法对原始光谱进行预处理,对全波段下的原始光谱和预处理后的光谱建立偏最小二乘回归(PLSR)模型,优选出的最佳预处理算法为去趋势校正法。原始数据经去趋势校正法预处理后,采用相关系数法选取特征波长,建立特征波长下羊肉含水率的 PLSR模型和逐步多元线性回归(SMLR)模型。结果表明,SMLR模型对含水率预测效果最好,校正集相关系数R_c为0.8597,标准误差SEC为0.0521;预测集相关系数R_p为0.8654,标准误差SEP为0.0387。研究表明,利用高光谱成像技术检测羊肉含水率是可行的。

Nondestructive detection of mutton moisture content based on hyperspectral technique

Nondestructive detection of mutton moisture content was studied by hyperspectral imaging system(1000~2500 nm). Spectral information was collected from 108 mutton samples. The original spectra were pretreated by standard normal variate,normalization,detrend correction,Savitzky-Golay,derivative and multiple scattering correction. The partial least squares regression(PLSR)model was established for original and pre-processed spectra in the whole band. The best preprocessing algorithm was the detrend correction method. The original data was pretreated by the detrend correction method,and the characteristic wavelength was selected by the correlation coefficient method. The PLSR and stepwise multiple linear regression(SMLR)model for the mutton moisture content were established. The results showed that the SMLR model had the best effect on the prediction of moisture content. The correlation coefficient(R_c)and standard error(SEC)for calibration set were 0.8597 and 0.0521 respectively,and correlation coefficient(R_p)and standard error(SEP)for predictive set were 0.8654 and 0.0387 respectively. The results showed that it is feasible to detect mutton moisture content by hyperspectral imaging technology.