基于光学成像的不同产地宁杞7号枸杞外观识别研究

Appearance Recognition of Ningqi No.7 Lycium barbarum from Different Geographical Origins Based on Hyperspectral Imaging

枸杞产地的快速、准确鉴别,对规范枸杞交易市场、推动不同产地枸杞差异化、品牌化战略发展具有重要意义。以2018—2019年宁夏回族自治区(简称宁夏)和新疆维吾尔自治区(简称新疆)4个种植区种植的宁杞7号夏果干果为研究对象,利用可见-近红外高光谱成像系统,对图像进行HSV(Hue,Saturation,Value)色彩空间变换和纹理特征提取。配合人工测定的百粒质量、果形指数(L/D)等枸杞果形参数指标,采用最小显著差异法(LSD)比较不同产地的差异性。在R 3.6.2环境支持下,对数据进行决策树(Decision Tree,DT)、随机森林(Random Forest,RF)、支持向量机(Support Vector Machine,SVM)、多元逻辑回归(Multinomial Logistic Regression,MLR)等分类器模型训练并建立产地识别模型,开展基于高光谱成像+计算机视觉的枸杞产地识别技术研究。结果表明,百粒质量新疆枸杞明显高于宁夏枸杞,果形指数宁夏枸杞高于新疆枸杞,宁夏枸杞纹理更深、更复杂,但色泽较暗;4种枸杞产地识别模型中DT模型表现最稳定,果形指数参与建模后的识别精度更高。

Rapid and accurate identification of origin of Lycium barbarum is of great significance for regulating trading market of Lycium barbarum and promoting differentiation and branding strategy development of Lycium barbarum in different origins.Dried fruit of Ningqi No.7 from four producing areas in Ningxia and Xinjiang from 2018 to 2019 was used as research object.Research object was scanned and imaged using visible-near-infrared hyperspectral imaging system.HSV color space transformation and texture feature extraction on scanned images were performed.Least significant difference method was used to compare differences between different origins in combination with manually measured 100-grain weight,fruit shape index and other Lycium barbarum fruit shape parameters.Data were trained with classifier models such as Decision Tree,Random Forest,Support Vector Machine,and Multinomials Logistic Regression,an origin identification model was established,and research on origin identification technology of Lycium barbarum was conducted.Results showed that the 100-grain weight of Lycium barbarum in Xinjiang was significantly higher than that in Ningxia,fruit shape index in Ningxia was better than that in Xinjiang,and texture of Lycium barbarum in Ningxia was deeper and more complex,but color was darker.Among four types of Lycium barbarum origin identification models,Decision Tree model has the most stable performance,and identification accuracy was higher after fruit shape index was involved in modeling.