利用波长范围在833～2 500 nm的傅里叶变换近红外光谱(Fourier transform near infrared spectroscopy, FT-NIR)对不同霉变程度的玉米颗粒进行检测区分。 首先, 为避免光谱数据首尾噪声影响, 对比四种常见的预处理方法, 最终选择移动平均平滑法对原始光谱数据进行预处理; 然后为选出合适的样本集划分方法以提高模型预测性能, 对常见的四种方法进行对比, 最终利用SPXY(sample set partitioning based on joint x-y distance)法进行样本集划分; 进一步为减少数据量, 降低维度, 使用连续投影算法(successive projections algorithm, SPA)提取出7个特征波长, 分别为833, 927, 1 208, 1 337, 1 454, 1 861和2 280 nm; 最后, 将七个特征波长数据作为输入, 选取径向基函数(radial basis function, RBF)作为支持向量机(support vector machine, SVM)核函数, 取参数C=7 760 469, γ=0.017 003建立判别模型。 SVM模型对训练集和测试集的预测准确率分别达到97.78%和93.33%。 另取不同品种的玉米颗粒, 以同样的标准挑选样品组成独立验证集, 所建立的判别模型对独立验证集的预测准确率达到91.11%。 结果表明基于SPA和SVM能有效地对玉米颗粒霉变程度进行判别, 所选取的7个特征波长为实现在线霉变玉米颗粒近红外检测提供了理论依据。

The feasibility of Fourier transform near infrared (FT-NIR) spectroscopy with spectral range between 833 and 2 500 nm to detect the moldy corn kernels with different levels of mildew was verified in this paper. Firstly, to avoid the influence of noise, moving average smoothing was used for spectral data preprocessing after four common pretreatment methods were compared. Then to improve the prediction performance of the model, SPXY (sample set partitioning based on joint x-y distance) was selected and used for sample set partition. Furthermore, in order to reduce the dimensions of the original spectral data, successive projection algorithm (SPA) was adopted and ultimately 7 characteristic wavelengths were extracted, the characteristic wavelengths were 833, 927, 1 208, 1 337, 1 454, 1 861, 2 280 nm. The experimental results showed when the spectrum data of the 7 characteristic wavelengths were taken as the input of SVM, the radial basic function (RBF) used as the kernel function, and kernel parameter C=7 760 469, γ=0.017 003, the classification accuracies of the established SVM model were 97.78% and 93.33% for the training and testing sets respectively. In addition, the independent validation set was selected in the same standard, and used to verify the model. At last, the classification accuracy of 91.11% for the independent validation set was achieved. The result indicated that it is feasible to identify and classify different degree of moldy corn grain kernels using SPA and SVM, and characteristic wavelengths selected by SPA in this paper also lay a foundation for the online NIR detection of mildew corn kernels.