高危阿片类毒品海洛因的泛滥, 对国家安定、 社会经济和人民生命财产安全带来了巨大的危害。 高效、 准确的海洛因及其代谢物的检测鉴定方法在打击毒品犯罪, 处理涉毒案件以及公安禁毒工作中具有十分重要的意义。 表面增强拉曼光谱（SERS）兼具检测速度快、 操作简便、 灵敏度高、 指纹识别及无损检测等优点, 能够实现对毒品的高效、 便携检测。 若结合模式识别技术, 可提高数据处理效率、 避免人为错判的发生, 进而实现自动精确分类识别的目的。 针对溶液中微/痕量海洛因及其代谢物, 提出了基于Au/SiO2复合纳米球阵列（Au/SiO2 NSA）的SERS检测与模式识别相结合的方法, 实现对它们的灵敏检测与高效鉴别。 首先, 采用气-液界面自组装和磁控溅射沉积的方法制备了具有良好SERS活性和结构一致性的Au/SiO2 NSA, 以此为SERS基底（芯片）, 结合便携式拉曼光谱仪, 成功实现了对水溶液中海洛因及其主要活性代谢物（6-单乙酰吗啡（6-MAM）和吗啡）的灵敏检测, 检测限低至10-4 mg·mL-1。 然后, 利用模式识别技术中的系统聚类分析（HCA）、 主成分分析（PCA）和支持向量机（SVM）对所获得的谱图数据进行定性/定量分类识别。 结果表明, 在HCA和PCA均能准确分类的基础上, 采用基于径向核函数、 线性核函数、 多项式核函数、 S型核函数中任意一种建立的PCA-SVM模型, 均能够100%地对海洛因、 6-MAM和吗啡进行定性识别； 选取基于径向核函数的SVM模型, 对不同浓度海洛因定量区分的准确率可达90.1%； 而通过基于线性核函数的SVM模型, 对不同浓度6-MAM和吗啡的判别准确率分别为84.8%、 70.2%。 这项工作不仅为基于SERS的灵敏检测与精准鉴别提供了一种具有实用价值的高质量基底（芯片）, 也为对海洛因及其代谢物进行准确的分类及识别给出了可行的方案。

The spread of high-risk opioid heroin has caused severe harm to national stability, social economy, and peoples life and property safety. Efficient and accurate detection/identification methods for heroin and its metabolites are significant in combating drug crimes, dealing with drug-related cases and anti-drug campaigns. Surface-enhanced Raman spectroscopy (SERS) has the merits of fast detection speed, simple operation, high sensitivity, fingerprint identification and non-destructivity, which can realize efficient and portable detection of drugs. If combined with pattern recognition, it can improve the efficiency of data processing and avoid the occurrence of human misjudgment. Thereby the purpose of automatic and accurate classification and identification can be achieved. In this work, to achieve sensitive detection and efficient identification of trace heroin and its metabolites in solution, a method combining SERS measurement based on Au-coated SiO2 composite nanosphere array (Au/SiO2 NSA) and pattern recognition is proposed. Firstly, gas-liquid interface self-assembly and magnetron sputtering deposition prepare Au/SiO2 NSA with good SERS activity and signal reproducibility. Employing such an array as SERS substrate (chip), combined with a portable Raman spectrometer, the high-efficiency detection of heroin and its main active metabolites (6-monoacetylmorphine (6-MAM) and morphine) in water solution is successfully achieved with the detection limit of 10-4 mg·mL-1. Next, to perform qualitative/quantitative identification of heroin and its metabolites, SERS spectral data are processed via hierarchical cluster analysis (HCA), principal component analysis (PCA) and support vector machine (SVM). When classifying heroin, 6-MAM and morphine, on the foundation of the accurate classification of them by both HCA and PCA, the PCA-SVM models based on radial basis function, linear kernel function, polynomial kernel function or sigmoid kernel function all can 100% qualitatively identify them. When adopting the PCA-SVM model to analyze heroin. 6-MAN and morphine quantitatively, the accuracy of quantitatively distinguishing different concentrations of heroin can reach 90.1% using the SVM model based on radial basis function. Via the SVM model of the linear kernel function, the accuracies in discriminating different concentrations of 6-MAM and morphine are 84.8% and 70.2%, respectively. This work provides a high-quality substrate (chip) with practical value for sensitive detection and accurate identification based on SERS but also puts forward a feasible approach for efficient classification and identification of heroin and its metabolites.