基于稀疏主成分分析的近红外光谱法鉴别黄花梨的成熟度

NIRS Identification of Maturity of Huanghua Pears with Sparse Principal Component Analysis

从同一果园的12棵果树上,在8月的4个不同日期各采集5个黄花梨样品,共采集240个样品。从每个样品上采集光谱数据。通过稀疏主成分分析(SPCA)和主成分分析(PCA)提取光谱中与成熟度相关的特征并进行解析,结合人工神经网络(ANN)建立黄花梨成熟度的鉴别模型。从所得载荷向量图可知:1 SPCA能有效提取光谱中与成熟度有关的特征,其7个稀疏主成分分别反映了黄花梨的糖类物质、水分、色素和硬度等信息;2 SPCA-ANN的成熟度鉴别模型的预测总识别率为93.33%,高于PCA-ANN的鉴别模型的预测总识别率91.67%。

Five Huanghua pears were collected from each of 12 pear trees in a same fruit yard in each of 4 definite dates in August, giving totally 240 pear samples. Spectral data was collected with each sample. Characteristic informations related to maturity in spectra were extracted by SPCA and PCA, and explanations were made. Discriminant models for maturity of the Huanghua pears were built by SPCA and PCA in combination with artificial neural network （ANN）. As shown in the loading vector diagrams, it was found that： ① SPCA was effective to extract specific characteristics related to maturity from the spectra. The seven sparse principle components were found to reflect separately informations about saccharides content, moisture content, pigmentation, hardness and so on of pears; ② Total recognition in prediction of maturity by models built by SPCA- ANN were attained to 93.33%, which is higher than 91.67% the total recognition attained by PCA-ANN.