主成分分析排序和模糊线性判别分析的生菜近红外光谱分类

贮存时间是影响生菜品质的一项重要因素，传统的贮存时间鉴别方法主要依靠人工经验，但是这种方法的准确率和可信度并不高。研究的目标是建立一种基于模糊识别的模型进行生菜光谱分析以实现生菜贮存时间的鉴别，并与其他鉴别方法作比较。为此，在当地超市购买60份新鲜生菜样品，存放于冰箱中待用。首先，通过Antaris Ⅱ近红外光谱检测仪采集生菜样品的近红外光谱数据，每隔12小时检测一次，每个样本检测重复三次，并取三次平均值作为实验数据。其次，利用多元散射校正（MSC）减少近红外光谱中的冗余信息。为了进一步去除近红外光谱中的无用信息以及简化随后的数据分类过程，分别运用主成分分析（PCA）和排序主成分分析 (PCA Sort)。其中，PCA Sort通过改进对主成分的排序方法能提高分类准确率，同时便于模糊线性鉴别分析（FLDA）进一步提取特征。PCA和PCA Sort的计算仅运用了前15个主成分（能充分反映光谱的主要信息）。最后，利用模糊线性鉴别分析算法（FLDA）和K近邻算法（KNN）进一步分类所得的低维数据。基于PCA和KNN算法的模型鉴别准确率达到43%，而基于PCA，FLDA和KNN算法的模型鉴别准确率可达83%。上述结果说明基于PCA，FLDA和KNN算法的模型鉴别准确率已经得到较大程度提高。当用PCA Sort替代了模型中的PCA算法后，结合FLDA和KNN算法则鉴别准确率达到98.33%。实验结果表明PCA Sort结合FLDA和KNN所建立的模型是有效的生菜贮存时间鉴别模型。

NIR Spectral Classification of Lettuce Using Principal Component Analysis Sort and Fuzzy Linear Discriminant Analysis

The storage time of lettuce is an important factor affecting the quality. The traditional way of detecting lettuce storage time mostly depends on artificial experience, so it lacks accuracy and reliability. This study aims to provide a fuzzy recognition model for spectral analysis of lettuce to identify the storage time of lettuce compared with other discriminant methods. For this objective, sixty samples of fresh lettuce bought in the local supermarket were prepared and stored in a refrigerator for later detection. These samples were detected by near-infrared spectroscopy (NIR). Firstly, the Antaris II NIR spectrometer (the wave number range: 10 000~4 000 cm-1) was utilized to collect the near-infrared spectral data of lettuce samples every 12 hours, and every sample detection was repeated three times, taking the average value as experiment data. Secondly,NIR spectra were preprocessed with multiple scatter correction (MSC) for decreasing reductant information. PCA and PCA Sort were used to further clear the useless data of NIR spectra and simplify the following classification of data. PCA Sort was based on PCA with sorting principal components and could improve the classification accuracy and help the FLDA extract features effectively. In this step, only the first fifteen components of PCA and PCA Sort were used to compress NIR spectra. Finally, fuzzy linear discriminant analysis (FLDA) algorithm and k-nearest neighbor (KNN) were performed to classify the previous low-dimensional data. The classification accuracy of the model based on PCA coupled with KNN was 43%, and that based on PCA as well as FLDA and KNN was 83%. The classification results in experiments showed that the discriminant of the model based on PCA, FLDA and KNN was significantly improved. Replacing PCA in the model with PCA Sort, the recognition accuracy of this new model based on the algorithm PCA Sortcoupled with FLDA and KNN was better and achieved 98.33%, which was higher than other classification algorithms. The classification results in experiments showed that PCA Sort plus FLDA and KNN could build an efficient discrimination model for the identification of the storage time of lettuce.