生菜的储藏时间是影响生菜新鲜程度的重要因素。 为了快速、 无损和有效地鉴别生菜的储藏时间, 以欧式距离的p次方代替模糊K调和均值聚类(FKHM)中欧式距离的平方提出了一种广义模糊K调和均值聚类(GFKHM)算法并将该算法应用于鉴别生菜的储藏时间。 以60个新鲜生菜样本为研究对象, 采用Antaris Ⅱ近红外光谱分析仪每隔12 h检测生菜的近红外漫反射光谱, 共检测三次, 光谱扫描的波数范围为10 000~4 000 cm-1。 首先用主成分分析(PCA)对1 557维的生菜近红外光谱进行降维处理以减少冗余信息, 取前20个主成分, 经过PCA处理后得到20维的数据。 然后用线性判别分析(LDA)提取光谱数据的鉴别信息以提高聚类的准确率, 取鉴别向量数为2, 则LDA将20维的数据转换为2维数据。 最后以模糊C-均值聚类(FCM)的类中心作为FKHM和GFKHM的初始聚类中心, 分别运行FKHM和GFKHM计算模糊隶属度以实现生菜储藏时间的鉴别。 结果表明, GFKHM的鉴别准确率能达到92.5%, FKHM的鉴别准确率为90.0%, GFKHM具有比FKHM更高的鉴别准确率。 GFKHM的聚类中心比FKHM更逼近真实类中心。 GFKHM的收敛速度明显快于FKHM。 采用近红外光谱技术同时结合GFKHM, PCA和LDA为快速和无损地鉴别生菜储藏时间提供了一种新的方法。

Lettuce storage time is an important factor affecting the lettuce freshness. To realize the non-destructive, rapid and effective discrimination of lettuce storage time, generalized fuzzy K-harmonic means (GFKHM) clustering was proposed by introducing the pth power of Euclidean distance into fuzzy K-harmonic means (FKHM) clustering to replace the square of Euclidean distance in FKHM, and furthermore GFKHM was applied in the discrimination of lettuce storage time. Sixty fresh lettuce samples were prepared as the research object, and the near infrared reflectance (NIR) spectra of lettuces were collected by Antaris Ⅱ near infrared spectrometer with a spectral range of 10 000～4 000 cm-1 for three 12-hour detections. Firstly, the 1 557-dimensional NIR spectra were reduced by principal component analysis (PCA) to decrease redundant information. After the first 20 principal components were selected, PCA translated the 1 557-dimensional NIR spectra into the 20-dimensional data. Secondly, linear discriminant analysis (LDA) was used to extract the discriminant information from the 20-dimensional data to improve the clustering accuracy. With the first two discriminant vectors, LDA translated the 20-dimensional data into the two-dimensional data. Finally, the cluster centers from fuzzy C-means clustering (FCM) were set as the initial cluster centers for FKHM and GFKHM and fuzzy membership values of FKHM and GFKHM were calculated to identify lettuce storage time. The experimental results showed that the discrimination accuracy of GFKHM has achieved 92.5% which was higher than that of FKHM. The cluster centers of GFKHM were much closer to the true cluster centers in comparison with FKHM. Furthermore, the convergence of the GFKHM was significantly faster than FKHM. Near infrared spectroscopy coupled with GFKHM, PCA and LDA could cluster NIR spectra of lettuce quickly and correctly, and this provided a fast and nondestructive method for identifying lettuce storage time.