基于高光谱技术的羊肉含水率测定方法研究

发布时间：2018-10-08 07:45

【摘要】：根据国家规定的羊肉卫生标准,含水率是评价羊肉品质的重要参考指标。本研究以内蒙古锡林郭勒羊肉为研究对象,利用高光谱成像系统分别在400nm-1000mn波段和1000nm-2500mn波段对108个羊肉样本进行光谱信息采集,并采用烘干法测定羊肉含水率。通过采集样品高光谱、光谱预处理、提取敏感波段,筛选含水率检测光谱特征参数,构建基于高光谱的含水率预测模型,并对模型进行了分析评价。主要研究内容和结果如下:1.设计光源照射角度调节装置,使用Solidworks对装置受力情况进行校核,并对运动轨迹进行三维仿真,验证该装置能够调节光谱仪光源照射角度。2.采集样本高光谱信息,通过多元散射校正等多种方法对原始光谱进行预处理,依据PLSR法模型优选400nm-1 OOOnm波段和1000nm-2500nm波段光谱预处理算法。结果表明400nm-1000nm波段和1000nm-2500nm波段的最佳预处理算法分别为标准正态结合去趋势化法和去趋势算法。3.采用PLS回归权重法分析光谱预处理后的数据,发现400nm-1000nm波段和1000nm-2500nm 波段敏感波长分别为 405.6nm、516.5nm、563.7nm、615.9mn、742.2nm、864.4nm、964.4nm 和 11OOmn、1346nm、1535nm、1635nm、1786nm、2111nm。4.分别采用偏最小二乘法和逐步多元线性回归法建立全波段和特征波段下的羊肉含水率预测模型。结果表明:特征波段下逐步多元线性回归模型预测效果优于偏最小二乘法模型,预测集模型相关系数Rp分别为0.8184和0.7984,标准偏差SEP分别为0.0581和0.0603;验证集模型相关系数Rc分别为0.8301和0.8231,标准偏差 SEC 分别为 0.0549 和 0.0587。本课题基于高光谱技术对羊肉含水率测定方法进行研究,避免了传统检测方法破坏样本的缺点,为日后便携式羊肉含水率检测仪的设计和开发提供了理论依据。
[Abstract]:According to the hygienic standard of mutton, moisture content is an important reference index to evaluate mutton quality. In this study, 108 mutton samples were collected in 400nm-1000mn band and 1000nm-2500mn band by hyperspectral imaging system, and moisture content of mutton was determined by drying method. By collecting sample hyperspectral, spectral pretreatment, extracting sensitive bands, screening moisture content to detect spectral characteristic parameters, a prediction model of moisture content based on hyperspectral was constructed, and the model was analyzed and evaluated. The main contents and results are as follows: 1. The light source angle adjusting device was designed, and the force condition of the device was checked by Solidworks. The 3D simulation of the motion track was carried out to verify that the device could adjust the illumination angle of the spectrometer light source. The sample hyperspectral information was collected and the original spectrum was pretreated by multiple scattering correction. The 400nm-1 OOOnm and 1000nm-2500nm spectral pretreatment algorithms were selected according to the PLSR model. The results show that the optimal preprocessing algorithms for 400nm-1000nm band and 1000nm-2500nm band are standard normal combination de-trend method and de-trend algorithm .3respectively. PLS regression weight method was used to analyze the data of spectral pretreatment. It was found that the sensitive wavelengths of 400nm-1000nm band and 1000nm-2500nm band were 405.6 nm ~ 516.5nm ~ (-1) ~ 563.7nm ~ (-1) ~ (615.9) mm ~ (-1) ~ (?) ~ 864.4nm ~ (-1) ~ (964.4) nm ~ (-1) ~ 1346nm ~ 1535nm ~ (1635nm) ~ (1635nm) ~ (1786) nmm ~ (-1) ~ (-1) nm ~ (-1) ~ (-1) nm. The partial least square method and stepwise multivariate linear regression method were used to establish the prediction models of mutton moisture content in the whole band and the characteristic band, respectively. The results show that the prediction effect of stepwise multivariate linear regression model is better than that of partial least square model. The correlation coefficient (Rp) of predictive set model is 0.8184 and 0.7984, the standard deviation SEP is 0.0581 and 0.0603, the correlation coefficient of verification set model is 0.8301 and 0.8231, and the standard deviation SEC is 0.0549 and 0.0587 respectively. Based on the hyperspectral technology, the method of moisture content measurement of mutton is studied in this paper, which avoids the shortcoming of the traditional method to destroy the sample, and provides a theoretical basis for the design and development of the portable mutton moisture content detector in the future.
【学位授予单位】：内蒙古农业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：O657.3;TS251.53

【参考文献】