针对以往卷烟产品质量监控效率低、时间长、缺少可视化结果展示等问题, 将近红外光谱分析技术应用到产品稳定性控制中, 提出了一种指纹图谱特征提取新方法。首先将卷烟近红外光谱进行主成分降维, 并以雷达图可视化形式描述产品质量稳定性趋势, 保证了图谱数据整体性与模糊性的统一。进而对包含样本全局信息的图形重心特征进行提取, 实现了对样本间内部结构的展示, 并建立了质量稳定性及异常类型的判别模型。实验结果表明, 和其他方法相比, 该特征提取方法取得了更好的识别效果, 可用于卷烟产品质量监控和真伪鉴别中。

Due to the low efficiency, long detection time, and lack of visual results show of the quality monitoring of cigarette products, the near infrared spectroscopy analysis techniques were applied to the process of product stability control and a new feature extraction method of the fingerprint spectrum was presented. Firstly, the principal component analysis method was used for the dimension reduction of the near infrared (NIR) spectra, and the radar chart was employed for the visual description of the quality stability trends of the products. This ensures the unification of integrity and fuzziness of the spectra data. And then, the barycentre features of the radar chart were extracted, which realized the demonstration of the internal structure of the samples. Also the model of the quality stability and the abnormal type was established. The study shows that this feature extraction method achieves desired result and can be used in the quality monitoring and counterfeit identification of the cigarette products.