苹果霉心病光谱在线检测的摆放姿态及建模方法优化研究

苹果营养丰富、口味酸甜，是深受大众喜爱的一种水果。苹果霉心病是一种真菌侵染果实病害，隐蔽性极强，一般在近成熟期果实内部发生霉变，肉眼从外观观察难以分辨，市面上大多数品种的苹果都受其影响。霉心病病果重量变轻、口感变差，严重的甚至不能食用，对经济效益的影响巨大。采用可见近红外光谱分析技术，使用微型光谱仪在线无损检测苹果霉心病，针对4种苹果在线输送时摆放姿态（竖放柄朝上、竖放柄朝下、横放柄朝输送方向和横放柄垂直输送方向）的判别效果进行了优化分析。首先使用主成分分析对600～900 nm波段的透射光谱提取主成分后分别建立线性判别分析（LDA）、马氏距离（MD）和K近邻法（KNN）模型并对校正集和预测集的判别准确率进行对比；其次对600～900 nm波段中心化预处理后建立偏最小二乘判别分析（PLS-DA）模型并给出4种摆放姿态的判别效果；最后使用两种机器学习算法极限学习机（ELM）和支持向量机（SVM）分别建立霉心病判别模型进行预测。对比上述所有6种判别模型，通过观察4种摆放姿态整体的判别效果得到最佳的建模方法为PLS-DA，其中竖放柄朝上和竖放柄朝下摆放的判别准确率都为93.75%，其他2种摆放姿态的判别准确率也都超过85%，再根据PLS-DA模型波段变量投影重要性指标得分值分布提取特征波段690～720 nm重新建立模型，对比4种摆放姿态效果最好的是竖放柄朝上摆放，其预测集的判别准确率达到93.75%，并且对病果的判别效果最佳。研究结果表明PLS-DA可以作为判别苹果霉心病一种有效方法，竖放柄朝上摆放可以作为苹果霉心病在线检测时一种有效姿态。

Optimization of Fruit Pose and Modeling Method for Online Spectral Detection of Apple Moldy Core

Moldy core of apples is a fungal disease that affects many commercially popular cultivars of apples.It is difficult to distinguish moldy core of the fruit from its appearance until the fruit is cut open. The objective of this study was to detect moldy core of apples by visible near-infrared spectroscopy (NIRS). The discrimination effects of four kinds of apple on-line transportation postures were compared: the apple stem upward, the apple stem downward, the apple stem towards the transportation direction, and the apple stem perpendicular to the transportation direction. Principal component analysis (PCA) was used to extract the principal components from the transmission spectra of 600~900 nm, and then linear discriminant analysis (LDA), Mahalanobis distance (MD) and k-nearest neighbor (KNN) models were established for comparison. The partial least squares discriminant analysis (PLS-DA) model was established after the central pretreatment of 600~900 nm. Two machine learning algorithms, extreme learning machine (ELM) and support vector machine (SVM)were also used to predict moldy core of apples. The best modeling method is PLS-DA. The accuracy rate of stem upward and stem downward was 93.75%, and the accuracy of the other two postures were more than 85%. Then according to VIP (variable importance in projection) scores, the characteristic band 690~720 nm was extracted, and the model was rebuilt. The best result of the four postures was apple stem upward. The accuracy rate of the prediction set was 93.75%.The results showed that PLS-DA could be used as an effective method to distinguish moldy core of apples, and the stem upward can be used as an effective posture for on-line detection of moldy core of apples.