为了使用快速、无损的方法区分激光打印文件使用的墨粉种类, 利用高光谱成像技术结合化学计量法对6种激光打印墨粉的光谱数据进行建模和种类鉴别的研究。利用可见-近红外高光谱成像仪采集400~1 000 nm波段内的光谱数据, 采用Savitzky Golay 平滑、标准化、多元散射校正和标准正态变量变换4种方法分别对光谱数据进行预处理, 而后分别建立随机森林(RF)、K最近邻(KNN)、支持向量机(SVM)、偏最小二乘判别分析(PLS-DA)和簇类独立软模式(SIMCA)模型, 进而实现激光打印墨粉的种类鉴别。利用准确率、拒识率和误识率3个指标作为模型评价标准。实验结果显示, SVM和PLS-DA模型的效果最佳, 准确率为100%, 拒识率和误识率为0。基于可见-近红外高光谱成像技术可以实现激光打印墨粉的快速种类鉴别。

In order to develop rapid and non-destructive method for identification of laser printer toner, six kinds of black toner were identified rapidly by combining hyperspectral imaging technique and five kinds of statistical learning method. Method: a visible and near-infrared hyperspectral imaging system covering the spectral range of 400-1 000 nm was set up to capture hyperspectral images of toner samples. Savitzky Golay smooth, normalize, multiple scatter correction and standard normal varite were applied as preprocessing method. After that, five statistical learning methods, including Random Forest (RF), K-nearest Neighbor (KNN), Support Vector Machine (SVM), Partial Least Square-discriminant analysis (PLS-DA) and Soft Independent Modeling of Class Analogy (SIMCA) were applied to establishment of discriminant models based on the full spectra. The properties of discriminant models were compared and valued by three parameters, precision, false reject rate (FRR) and false accept rate (FAR). Result: Among all discriminant models, the SVM and PLS-DA model show the best identification result, the precision is 100%, FRR and FAR are both 0. Conclusion: black toner could be identified by visible and near-infrared hyperspectral imaging technique combined with statistical learning method rapidly.