摘要
传统的颗粒饲料淀粉含量检测方法具有准确度低、破坏样本等弊端。因此,文章研究了基于智能图像颗粒饲料淀粉含量检测方法,采用近红外光谱检测玉米籽粒内直链淀粉含量,分析玉米籽粒淀粉含量检测的特征波长优选方法,在不同频波段区间中,通过相关系数法以及遗传算法优选近红外光谱建模变量,对玉米籽粒直链淀粉含量的定量分析模型进行构建,得到最优的淀粉含量波段为是波长(3 999～4 157+5 152～5 210 cm-1),最佳PLS隐变量数是2个,最佳预处理方法是BASELINE+MC方法,最佳LS-SVM模型的RMSEP是0.027。将玉米透明度感官评分当成因变量,将图像透明度特征值当成自变量,通过逐步融入-过滤法实施多元线性回归分析,过滤出干扰玉米透明度的4个关键特征值(R值、B值、V值以及M值),塑造玉米透明度预测模型。通过双波长法检测玉米内直链以及支链淀粉含量和二者间的比例,研究透明度特征值同淀粉含量的相关性,以玉米直链淀粉含量为因变量,透明度特征值(M,h)为自变量,实施多元线性回归分析,构建玉米淀粉含量检测模型,实现玉米淀粉含量的准确检测。
The traditional method of detecting starch content in pellet feed has disadvantages of low accuracy and destroying samples. Therefore, this article studies the detection method of intelligent image based on the content of starch grain feed, using near infrared spectroscopy in maize grain amylose content, analysis and optimization method of characteristic wavelength detection of maize grain starch content, in different frequency band interval, by correlation coefficient method and genetic algorithm optimization modeling of near infrared spectra for quantitative variables. Corn amylose content analysis model was constructed, getting the starch content for optimum wavelength bands (3 999-4 157 +5152.5210 cm^-1), the best PLS latent variable number is 2, the best pretreatment method is the BASELINE+MC method, the optimal model of LS-SVM RMSEP is 0.027. The corn transparency sensory score as dependent variable, the image feature values as independent variables through transparency gradually into the filter implementation of the law of multiple linear regression analysis to filter out the 4 key features of the transparency of the R value of corn interference and B value, V value and M value, shaping the corn transparency prediction model. The dual wavelength method in detection of maize amylose and amylopectin content and the ratio between the two, with the coiTelation between the starch content of the transparency of characteristic value of corn amylose content when the variable transparency feature values (M, H) as independent variables, the multivariate linear regression analysis, constructs the detection model of com starch and to achieve accurate detection of maize starch content.
出处
《饲料研究》
CAS
北大核心
2019年第2期53-57,共5页
Feed Research
关键词
智能
图像
颗粒
饲料
淀粉
含量
检测
intelligence
image
granule
feed
starch
content
detection
