论文部分内容阅读
为探讨高光谱成像技术无损检测马铃薯环腐病的可行性,采用反射高光谱(980~1 650 nm)成像技术,以120个马铃薯样本(合格60个,环腐60个)为研究对象,对比多元散射校正、标准正态变换、卷积+一阶导数等对建模的影响,优选出多元散射校正的光谱预处理方法;然后基于偏最小二乘回归系数法提取9个特征波长(993、1 005、1 009、1 031、1 112、1 162、1 165、1 225、1 636 nm),建立特征波长下马铃薯环腐病的2类线性判别分析(linear discriminant analysis,LDA)模型和4类支持向量机(support vector machine,SVM)模型,即Fisher-LDA、马氏距离-LDA、线性核SVM、径向基核SVM、多项式核SVM和S型核SVM。结果表明,LDA模型中马氏距离法最优,SVM模型中S型核SVM最优,LDA模型整体优于SVM模型,最终确定基于马氏距离LDA的马铃薯环腐病判别模型为最佳模型,校正集、验证集识别率分别为100%和93.33%。实验结果表明高光谱无损检测马铃薯环腐病具有可行性。
In order to investigate the feasibility of non-destructive detection of ring rot of potato by hyperspectral imaging, 120 potato samples (60 passions and 60 rotations) were studied by reflectance hyperspectral (980 ~ 1 650 nm) imaging technique, Multivariate scatter correction, normalized normal transform, convolution + first derivative and so on, the pretreatment method of multivariate scatter correction is optimized. Then nine characteristic wavelengths (993, 1 005, 009, 1 031, 1 112, 1 162, 1 165, 1 225 and 1 636 nm), two linear discriminant analysis (LDA) models of potato ring rot with characteristic wavelengths and four Support vector machine (SVM) models, namely Fisher-LDA, Mahalanobis-LDA, linear kernel SVM, radial kernel SVM, polynomial kernel SVM and S-kernel SVM. The results show that the Mahalanobis distance method is the best in the LDA model, the SVM in the SVM model is the best, and the LDA model is better than the SVM model in the whole. Finally, the best discriminant model of potato ring rot based on Mahalanobis distance LDA is determined, Calibration set, validation set recognition rate were 100% and 93.33%. The experimental results show that hyperspectral nondestructive detection of potato ring rot is feasible.