Discrimination of black gel pen inks by hyperspectral imaging combined with logistic regression

A rapid and non-destructive method for discriminating black gel pen inks was proposed by combining hyperspectral imaging (HSI) with logistic regression (LR) to address the challenges in differentiating black gel pen inks in document examination, as well as the limitations of conventional detection methods that cause damage to evidence samples. Hyperspectral data of handwriting samples from 22 brands (models) of black gel pens were collected in the range of 400-1000 nm using an HSI system, and 4400 mean spectra (200 per class, 121 bands) were extracted from regions of interest (ROIs). Among four candidate preprocessing schemes, Z-Score standardization was selected as the optimal one. LR, multilayer perceptron (MLP) and k-nearest neighbour (KNN) classifiers were constructed. The band contribution was further interpreted based on the LR weight matrix. The results showed that Z-Score standardization was the key preprocessing step, improving the 5-fold cross-validation accuracy of the LR model from 63.07% to 97.09%. The LR model achieved the best performance with a test accuracy of 97.50%, and the F1 scores of all 22 classes were no less than 0.935, while MLP and KNN reached 92.05% and 73.30%, respectively. The proposed method is rapid, non-destructive, accurate and interpretable, providing technical support for ink type identification in forensic document examination.