利用高光谱成像技术对轻度擦伤桃子按碰伤后时间进行监督分类

摘要：桃子对机械冲击、挤压等损伤较为敏感，损伤发生后，受损伤的桃子容易发酵、腐烂或发霉，并感染其他未受损 伤的桃子。适度的擦伤对消费者的购买欲望影响大于价格。对于早期擦伤的桃子，软化和褐变不会立即出现（2 天内）， 很难区分果实是否受损伤。利用高光谱成像装置采集正常桃子和碰撞后12、24、36、48h的数据，分析这5种类型桃子 的光谱特征，建立桃子碰撞后生理指标的PCA分类模型。提取变色区域和正常区域的光谱和图像特征。其次，将采集的 图像数据集按 3：1 的比例分为建模集和预测集，输入光谱和图像特征，建立桃子伤痕时间判别模型。结果表明：基于 PLS-DA算法，当输入变量为光谱特征时，桃子伤痕后12、24、36、48 h的分类正确率分别为96.67%、96.67%、93.33%、 83.33%，建模集的相关系数（（ Rc）达到0.928；基于LS-SVM算法，当输入变量为光谱特征时，核函数RBF_Kernel分类 正确率达到80%、96.67%、100%、100%。本研究表明高光谱特征技术可以识别桃子早期碰伤，为在线检测桃子短期碰伤 的预测提供理论依据。

关键词：高光谱成像技术 早期碰伤 桃子 判别模型 分类准确率

期刊：Infrared Physics and Technology

DOI：https://DOI.org/10.1016/j.infrared.2020.103557

设备：GaiaField

Supervised classification of slightly bruised peaches with respect to the time after bruising by using hyperspectral imaging technology

Abstract：Peaches are sensitive to bruising from mechanical impact and compression. Peaches affected by bruises will tend towards fermentation, decay or mildew and infect other non-bruised ones after damage occurrence. A moderate amount of bruising is a barrier to purchase desire of consumers instead of price. For the early bruised peaches, softening and browning does not appear immediately (within 2 days), and it is difficult to distinguish the problem of fruit bruises. The hypersecretion imaging device was used to collect the data of normal peaches and the data of 12, 24, 36 and 48 h after the encounter, and the spectral characteristics of these five types of peaches were analyzed to establish the PCA classification model of physiological indexes of peaches after the collision. Extract the spectrum and image features of the discolored area and the normal area. Secondly, divide the collection of image data sets into the modelling set and the prediction set at a ratio of 3:1. Input the spectrum and image feature to establish a time discrimination model for peach bruises. The results show that it is based on the PLS-DA algorithm When the input variable is a spectral feature, the classification accuracy rates at 12, 24, 36, and 48 h after bruising are 96.67%, 96.67%, 93.33%, and 83.33%, respectively, and the correlation coefficient (Rc) of the modelling set reaches 0.928; Based on LS -SVM algorithm, when the input variable is a spectral feature, the kernel function RBF_Kernel classifies correctly to 80%, 96.67%, 100%, 100%. This study shows that the spectral feature of hypersecretion technology can identify early bruises of peaches, which are online testing provide a theoretical basis for predicting short-term bumps of peaches.

Keywords：Hyperspectral imaging technology Early bruises Peaches Discriminant models Classification accuracy