高光谱成像技术鉴别菠菜叶片农药残留种类

利用高光谱成像技术无损鉴别菠菜叶片农药残留种类。采用高光谱成像仪采集900~1 700 nm波段内的光谱数据，采用多元散射校正对光谱数据进行预处理。利用主成分分析对不同种类菠菜样品的光谱数据进行分析，结果表明主成分分析能在可视化层面对不同种类的农药残留菠菜样品进行有效判别。另外，将卡方检验特征选择算法分别与支持向量机、朴素贝叶斯、决策树和线性判别分析算法结合，并采用10-fold交叉验证评价方法，筛选出最佳波段和最优判别模型（线性判别模型）。筛选出的8个特征波长为1 439.3，1 442.5，1 445.8，1 449，1 452.3，1 455.5，1 458.7，1 462 nm，模型的预测准确率达到0.993且10次交叉验证的标准差为0.009。结果表明，基于高光谱成像技术能准确地识别菠菜叶片上的农药残留种类。

Identification of Pesticide Residue Types in Spinach Leaves Based on Hyperspectral Imaging

Non-destructive identification of pesticide residues in spinach was studied using hyperspectral imaging. The hyperspectral images between 900 nm and 1 700 nm were obtained with the help of hyperspectral imager. The original spectra were corrected by multivariate scatter correction (MSC). The principal component analysis (PCA) was used to analyze the spectral data of different spinach samples, the results showed that PCA could effectively discriminate different kinds of pesticide residues spinach samples on the visualization level. In addition, chi-squared test feature selection algorithm was separately combined with four learning algorithms (e.g. support vector machine, naive Bayes, decision tree and linear discriminant analysis) to get the best bands and optimal discriminant model(linear discriminant model) with the help of 10-fold cross-validation technique. The selected eight characteristic wavelengths are 1 439.3, 1 442.5, 1 445.8, 1 449, 1 452.3, 1 455.5, 1 458.7, 1 462 nm and the prediction accuracy by optimal discriminant model is 0.993 and 10 times of cross validation standard deviation is 0.009. The results show that hyperspectral imaging technology can accurately identify the types of pesticide residues on spinach leaves.