用遗传算法( Genetic Algorithm，GA) 搜寻可识别被不同农药污染脐橙的可见/近红外光谱的最佳特征光

谱区间及波长，并建立了支持向量机( Support Vector Machines，SVM) 定性分析模型。实验供试农药为灭多威、

氰戊菊酯和氧乐果3 种。通过GA 来搜寻整个波段范围( 460 ～ 1 800 nm) ，将得到的9 个最佳特征光谱区间所

包含的波长( 共318 个) 作为SVM 建模的输入变量，对识别被3 种农药污染脐橙的准确率为100 。并继续应

用GA 优化，得到71 个特征波长，此时建立的SVM 模型的识别准确率为99． 57 。虽然识别的准确率有所下降，但是模型的复杂程度得到了很大的优化，其输入变量减少到71 个。实验结果表明利用可见/近红外光谱技

术结合SVM 方法可以有效识别被不同农药污染的脐橙。-Genetic algorithm ( GA) was used to search for the best characteristic spectral ranges andwavelengths of visible /near － infrared spectra ( Vis /NIRs) ，a qualitative analysis model of support vector machine

( SVM) was set up to recognize navel oranges contaminated with different pesticides． The pesticides in

the experiment were Methomyl，fenvalerate and omethoate． Using GA to search the entire band range ( 460 ～1 800 nm) ，the 9 best characteristic spectral ranges ( 318 wavelengths) were used as the input variables of SVM model and the accuracy of the prediction set classification was 100 ． Then GA method was used continually

and 71 wavelengths were extracted，the corresponding SVM model was built with 99． 57　accuracy． Although

the classification accuracy rate declined，the complexity of the model was greatly optimized by reducing

the input variables to 71． The experiment results showed that the application of Vis /NIRs combined with SVM can effectively detect the navel oranges conta