基于高光谱成像的苹果品种快速鉴别

Rapid Identification of Apple Varieties Based on Hyperspectral Imaging

摘要:

以“乔纳金”苹果，“红富士”苹果和“秦冠”苹果共90个试验样本为试材分别采集865～1711nm的近红外波段高光谱图像，选取苹果图像感兴趣区域（ROI），以分辨率2.8nm提取其平均反射光谱数据，分别利用K近邻法（KNN）和径向基核函数支持向量机（RBF-SVM）进行品种判别，5折交叉检验。结果表明，3种苹果的近红外高光谱图像均在波长941~1602nm之间变得清晰，该区域200个波段下的平均反射光谱数据经KNN法中的10种距离算法评判，当K取值3和5时，切比雪夫距离、欧几里得距离和明可夫斯基距离3种距离算法的识别正确率均达到100%；SVM-RBF核函数模型中，γ取值为2-8~1的范围内识别正确率均在92%以上，当γ取值2-5，C取值为16和32时，识别正确率最高，为96.67%。故利用近红外高光谱图像技术结合KNN计算对苹果品种进行快速鉴别是优异和可靠的方案。

Abstract:

In order to achieve rapid non-destructive identification of apple varieties, the methodology of near-infrared hyperspectral imaging on identification of apple varieties was investigated. Near infrared hyperspectral images with wavelength from 865~1711nm of total 90 sample fruits were collected from three different varieties (“Jonagold”, “Fuji” and “Qinguan” apples), and hyperspectral image area of the apple was selected as a region of interest (ROI). Reflection intensity data of the average reflex spectrum were extracted with resolution rate of 2.8nm, then they were calculated with K-nearest neighbor (KNN) and the support vector machine (SVM) methods, respectively, which were checked with 5-fold cross-validation method. The results showed that the hyperspectral images of three varieties of apples all became clear within wavelength of 941~1602nm. Among ten distance-types’ judgment of KNN with average reflection intensity at 200 wavelength-points, the identification accuracy of Chebychev, Euclidean and Minkowski reached the highest of 100% when the parameter K was set at 3 or 5. While using the support vector machine-radial basis function (SVM-RBF) model, the accuracy rate reached above 92% when the value of γ fell within 2-8～1. The highest recognition rate of this model reached 96.67% when γ was set at 2-5 and C took the value of 16 amd 32 at the same time. The results demonstrated that near-infrared hyperspectral imaging in combination with KNN was excellent and reliable for the rapid identification of apple varieties. This method could provide reference for identifying apple varieties in production.