目的 建立基于颜色气味数字化融合信息的苦杏仁Armeniacae Semen Amarum走油程度判别分析和内在质量预测模型。方法 基于颜色气味数字化的融合信息，联合机器学习算法对其走油程度进行判别，对比其正判率，寻找识别效果最好的算法。利用SPSS分析平台，开展苦杏仁颜色气味融合信息数字化与内在化学成分的相关性分析，同时建立含量预测回归方程并且检验其拟合度。结果 基于粉末颜色和气味融合信息的模型建立中，Logistic、IBK、K-Star、LMT和Random Forest算法正判率较高，可完成对不同走油程度苦杏仁的分类鉴别；基于剖面颜色和气味融合信息的模型建立中，Logistic算法和K-Star算法可完成走油程度的判定。基于粉末颜色及气味融合信息建立苦杏仁质量预测模型，预测方程：Y_苦杏仁苷＝2.175－1.340 F_1-2＋0.529 F_1-1，R²＝0.732；Y_酸值＝2.113＋1.724 9 F_1-2－0.667 F_1-1，R²＝0.719；基于剖面颜色及气味融合信息建立苦杏仁质量预测模型，预测方程如下：Y_苦杏仁苷＝2.153＋1.242 F_2-2＋0.5 F_2-1－0.689 F_2-3，R²＝0.775，Y_酸值＝2.226－1.946 F_2-2－0.785 F_2-1，R²＝0.738，拟合度结果优良。结论 通过颜色气味数字化信息可快速推断苦杏仁化学成分的变化趋势，颜色气味融合信息测量可发展为苦杏仁质量评价的新方法。

Objective To establish discriminative analysis and quality prediction models of Kuxingren (Armeniacae Semen Amarum, ASA) with different rancidness degrees based on fusion information of color and odor digital values. Methods Firstly different models were established by various classifiers to discriminate different rancidness degrees of ASA samples with positive judgment rate as evaluation index. And then the optimal algorithm was screened out based on fusion information of color and odor digital values. Using SPSS analysis platform, the correlation analysis between the digitalized odor and color fusion information and the chemical composition of ASA samples was carried out. Afterwards the content prediction regression equation was established and its fit was checked. Results In the model building based on the fusion information of powder color and odor, four kinds of algorithms, namely Logistic, IBK, K-Star, LMT and Random Forest, possess higher positive judgment rate and could complete the classification and identification of ASA samples with different rancidness degrees; In the model building based on the fusion information of longitudinal section color and odor, Logistic algorithm and K-Star algorithm could complete the determination of rancidness degrees. The prediction models of ASA quality were established based on the fusion information of powder color and odor, and the prediction equations were as follows:Y_amygdalin=2.175-1.340 F_1-2 + 0.529 F_1-1, R²=0.732; Y_{acid value}=2.113 + 1.724 9 F_1-2-0.667 F_1-1, R²=0.719; based on the fusion information of longitudinal section color and odor ASA quality prediction models were established with the following prediction equations:Y_amygdalin=2.153 + 1.242 F_2-2 + 0.5 F_2-1-0.689 F_2-3, R²=0.775, Y_{acid value}=2.226-1.946 F_2-2-0.785 F_2-1, R²=0.738, with excellent fit results. Conclusions The color-odor digital information was used to profile the trend of chemical composition of ASA samples, and the color-odor fusion information measurement could be used for the quality evaluation of ASA samples.

R283.6

国家自然科学基金面上项目（81573542）；国家自然科学基金青年科学基金项目（81403054）