为了快速无损检测不同品种（“松花”、“雪白”）花椰菜中硫代葡萄糖苷的含量。本实验采集新鲜收获的“松花”、“雪白”花椰菜样本进行可见/近红外光谱的采集、提取和分析。首先，采用基线校正（Baseline）、标准正态变量变换（Standard Nomal Variate transform，SNV）、中值滤波（MedianFilter，MF）、高斯滤波（GaussionFilter，GF）、S-G平滑（Savitzky-Golay）五种方法进行原始光谱的预处理分析。然后，分别采用连续投影算法（Successive Projections Algorithm，SPA）、回归系数法（Regression Coefficient，RC）进行特征波段的提取，并采用主成分分析法（Principal Component Analysis，PCA）进行主成分的提取，在此基础上，结合最佳预处理方法建立偏最小二乘回归（Partial Least Squares Regression，PLSR）模型。结果表明：“松花”花椰菜的光谱数据所建立的MF-PCA-PLS模型最佳，校正集模型参数Rc=0.89，RMSEC=1.23，预测集模型参数Rp=0.89，RMSEP=0.63。“雪白”花椰菜光谱数据所建的MF-RC-PLS模型最优，校正集模型参数Rc=0.87，RMSEC=1.31，预测集模型参数Rp=0.73，RMSEP=0.46。由此可见，近红外光谱结合PLSR算法能够快速、无损、准确地检测花椰菜中硫代葡萄糖苷的含量。

In order to quickly detect the content of glucosinolates in different varieties of cauliflower (“Songhua” and “Snow-white”), in this experiment, fresh "Songhua" and "Snow-white" cauliflower samples were collected for visible/near-infrared spectrum collection, extraction and analysis. First, five methods, Baseline correction, Standard Nomal Variate transform (SNV), Median Filter (MF), Gaussion Filter (GF) and S-G smoothing (savitzky-golay), were used for the preprocessing analysis of the original spectra. Secondly, methods of extracting feature bands were respectively adopted, namely the Successive projection Algorithm (SPA) and the Regression Coefficient (RC) were used to extract the feature bands. Principal Component Analysis (PCA) was used to extract the Principal components. A Partial Least Squares Regression (PLSR) model was established based on the optimal preprocessing method. The results showed that the MF-PCA-PLS model established by the spectral data of "Songhua" cauliflower was the best, and the calibration set model parameter Rc=0.89, RMSEC=1.23, predictive set model parameters Rp=0.89, RMSEP=0.63.The MF-RC-PLS model based on the spectral data of "Snow-white" cauliflower was optimal, with calibration set model parameters Rc=0.87, RMSEC=1.31, predictive set model parameters Rp=0.73, RMSEP=0.46. It can be seen that the content of glucoside in cauliflower can be detected quickly, nondestructively and accurately by PLSR algorithm combined with visible/near-infrared spectroscopy.

国家自然科学基金青年基金项目（31801632）；山西省高等学校科技创新项目（2019L0396）；山西农业大学科技创新基金项目（2016YJ04）