中文摘要：

为了实现褐变板栗的快速无损分选,研究了基于近红外光谱技术的褐变板栗栗仁检测方法。试验在1000-2500 nm波段范围内采集板栗栗仁的反射光谱,通过标准正态变量变换预处理后,采用K-最近邻法（KNN）、簇类独立软模式法（SIMCA）、主成分回归-线性判别分析法（PCA-LDA）、偏最小二乘回归-线性判别分析法（PLS-LDA）以及最小二乘-支持向量机判别分析法（LS-SVM）分别建立褐变板栗识别模型并进行比较分析。偏最小二乘结合最小二乘-支持向量机所建PLS-LS-SVM模型性能最优,该模型对测试集的敏感性、特异性和识别正确率分别为1.00%、0.92%和95.00%。结果表明：近红外光谱结合PLS-LS-SVM可用于褐变板栗的快速无损检测。

英文摘要：

Near-infrared spectroscopy technology was employed in the present study to detect chestnut browning in peeled chestnut rapidly and non-destructively. 70 normal chestnuts and 110 brown heart chestnuts were prepared and their diffuse reflectance spectrums were collected in the range of 1000-2500 nm. Standard normal variate（SNV） was used before further analysis. K-nearest neighbors method（KNN）, soft independent modeling of class analogy（SIMCA）, principal component analysislinear discriminant analysis（PCA-LDA）, partial least squares-linear discriminant analysis（PLS-LDA）, least squares-support vector machines（LS-SVM） modeling methods were built to detect browning chestnut. These modeling methods were evaluated and the best model was selected. The results showed that PLS-LS-SVM model outperformed KNN, SIMCA, PCA-LDA and PLS-LDA. The sensitivity, specificity and accuracy were 0.92%, 1.00% and 95.00% respectively. The results indicated that near-infrared spectroscopy combined with PLS-LS-SVM allows accurate detection of peeled chestnut browning, providing important reference for rapid determination of chestnut quality.