高光谱技术结合变量选择方法的甘薯冻害检测研究

Study on the detection of sweet potato freezing damage based on hyperspectral technology and variable selection method

为建立一种快速无损鉴别甘薯冻害的检测方法,利用高光谱技术采集900~1700 nm完好和冻害两类甘薯的高光谱信息并提取样本完好和冻害区域光谱,获得完好和冻害区域光谱343和476个。采用Kennard-Stone算法挑选训练集和预测集中完好和冻害样本。采用4种预处理方法对原始光谱预处理,选出一阶导数(first derivative,FD)为最佳的预处理方法。通过竞争性自适应重加权算法(competitive adaptive reweighted sampling,CARS)、迭代保留信息变量算法(iteratively retains informative variables,IRIV)以及结合连续投影算法(successive projections algorithm,SPA)的CARS-SPA和IRIV-SPA结合算法分别筛选出46、65、24和35个特征波长,并应用偏最小二乘法(partial least squares,PLS)和最小二乘支持向量机(least squares support vector machines,LS-SVM)建立甘薯冻害识别模型。结果表明,高光谱技术可以有效对甘薯冻害进行检测,CARS方法可以有效选择有用波长变量,是优于IRIV、CARS-SPA和IRIV-SPA算法的特征波长提取方法。CARS-PLS模型运算速度快且预测结果最优,其预测集样本的识别正确率、灵敏度及特异性分别为98.05%、98.84%和97.48%。该研究实现了对甘薯冻害特征的识别,为后续甘薯品质在线分选检测研究和在线检测设备的开发奠定了理论基础。

Freezing damage is an important factor that causes rot in sweet potatoes.Timely detection and elimination of freezing damage of sweet potato is the key to reducing production losses.To establish a rapid and non-destructive detection method for freezing damage of sweet potato in the early stage,this study established a recognition model for the freezing damage of sweet potato in the early stage based on hyperspectral technology and chemometric methods."Gaia Sorter",a kind of hyperspectral sorter,was used to collect the hyperspectral information of the intact and frost-damaged sweet potato samples in the 900-1700 nm band,and extract the average spectrum of the intact and frost-damaged regions of the samples to obtain 343 spectra for the intact region and 476 spectra for the frost-damaged region.The Kennard-Stone algorithm was used to select intact and frost-damaged samples in the training set and prediction set.Four preprocessing methods were used to preprocess the original spectra,and the first derivative(FD)was selected as the best preprocessing method.The competitive adaptive reweighted sampling(CARS),iteratively retains informative variables(IRIV),and CARS-SPA and IRIV-SPA combined with successive projections algorithm(SPA)were used to screen out 46,65,24 and 35 characteristic wavelength variables related to sweet potato freezing damage.Partial least squares(PLS)and least squares support vector machines(LS-SVM)were used to establish identification models of freezing damage in sweet potato,and it was compared with the recognition model without variable screening.The results showed that hyperspectral technology could effectively detect the freezing damage of sweet potatoes.The CARS method could effectively select useful wavelength variables.It was superior to the IRIV,CARS-SPA and IRIV-SPA algorithms.The CARS-PLS model showed fast calculation speed and the best prediction results.The recognition accuracy,sensitivity and specificity of samples in the prediction set were 98.05%,98.84%and 97.48%,respectively.This study has achieved the identification of the characteristics of freezing damage in sweet potato,which lays a theoretical foundation for the subsequent research on online sorting and detection and the development of online testing equipment for the sweet potato quality.