生菜的新鲜程度是影响生菜品质的最重要因素之一, 其主要取决于生菜的储藏时间, 因此, 对不同储藏时间的生菜进行准确鉴别具有重要研究价值。 由于不同储藏时间生菜的近红外光谱数据具有差异性的特点, 因而使用近红外为不同储藏时间的生菜进行鉴别分类是可行的。 通过将联合模糊C均值聚类(allied fuzzy c-means, AFCM)中的欧式距离测度替换为指数距离测度从而提出了一种GG联合模糊聚类(Gath-Geva AFCM, GGAFCM)分析算法。 GGAFCM通过迭代计算得到模糊隶属度值和典型值, 再结合近红外光谱实现了对不同存储时间生菜的高效精准鉴别。 以新鲜的生菜样本作为研究对象, 使用傅里叶近红外光谱仪(Antaris Ⅱ型)每隔12 h对生菜样本采集漫反射光谱数据, 光谱的波数范围介于10 000~4 000 cm-1之间。 首先, 通过主成分分析(principal component analysis, PCA)对采集到的1 557维生菜近红外光谱数据进行数据压缩将其降至22维, 然后通过模糊线性判别分析(fuzzy linear discriminant analysis, FLDA)对降维后的近红外漫反射光谱数据的鉴别信息进行提取。 设定鉴别向量数为2, 即通过FLDA将22维的生菜近红外光谱数据转换为了2维数据。 最后将模糊C均值聚类(fuzzy c-means, FCM)的聚类中心作为GGAFCM和AFCM的初始聚类中心, 通过运行FCM, GGAFCM和AFCM完成对不同储藏时间生菜的鉴别分类, 并对三种模糊聚类算法得到的聚类准确率、 模糊隶属度、 迭代次数进行分析。 实验结果表明: 在初始化条件相同的情况下, 采用的GGAFCM算法与FCM和AFCM算法相比具有更高的鉴别准确率。 在m=2的情况下, GGAFCM的鉴别准确率达到了95.56%, 而AFCM的聚类准确率为91.11%。 GGAFCM迭代4次达到收敛, 而AFCM与FCM均需要8次迭代计算才能达到收敛。 基于近红外光谱技术, 通过GGAFCM结合PCA与FLDA算法可以高效快速且无损的完成对储存时间不同的生菜的准确鉴别分类, 为生菜储存时间的准确、 快速鉴别提供了实验依据和参考方法, 具有一定的实际应用价值。

The freshness of lettuce is one of the most important factors affecting the lettuce quality, and it depends on the storage time. Therefore, it has important research value to identify the lettuce samples with different storage time accurately. Because the near-infrared reflectance (NIR) spectra of lettuce with different storage time have different characteristics, it is feasible to use NIR technology to identify lettuce with different storage time. Gath-Geva allied fuzzy c-means (GGAFCM) clustering was proposed to replacing the Euclidean distance in allied fuzzy c-means (AFCM) clustering with the exponential distance. By iterative computations, GGAFCM can produce fuzzy membership and typical values, which combine with near-infrared reflectance spectroscopy (NIRS) to achieve the classification of the lettuce samples with different storage time accurately. The experiment was conducted on fresh samples of lettuce, which were collected with Antaris Ⅱ spectrometer every 12 hours. The spectral wavenumber ranges from 10 000 to 4 000 cm-1. At first, by principal component analysis (PCA), the 1 557-dimensional spectra of lettuce samples were compressed to the 22-dimensional data whose discriminant information was extracted by fuzzy linear discriminant analysis (FLDA). As a result, the 22-dimensional data were transformed into the two-dimensional data by FLDA with two discriminant vectors. At last, the cluster centers of fuzzy c-means (FCM) clustering acted as the initial cluster centers of both GGAFCM and AFCM, and lettuce samples with different storage time were identified by FCM, GGAFCM and AFCM whose clustering accuracies, fuzzy membership values and iterative times were analyzed. Experimental results indicated that with the same initialization conditions, the GGAFCM algorithm adopted in this study has higher discrimination accuracy than FCM and AFCM. In the case of m=2, the discrimination accuracy of GGAFCM reached 95.56%, while the clustering accuracy of FCM and AFCM was 91.11%. GGAFCM converged after 4 iterations, while both AFCM and FCM needed 8 iterations to reach convergence. Based on NIRS, GGAFCM combined with PCA and FLDA can efficiently, quickly and nondestructively complete the accurate identification of lettuce samples with different storage time. It provides the experimental foundation and reference method for accurate and rapid identification of lettuce storage time and has certain practical application value.