Identification of writing marks from pencil lead through machine learning based on laser-induced breakdown spectroscopy

Forensic science indicates writing marks as a powerful source of information for identification of questioned documents in the judicial field. Accordingly, these marks should be analyzed, and whether the marks above originate from the identical or different sources should be verified. In this study, various pencil marks were taken as the examples to be identified through machine learning based on laser-induced breakdown spectroscopy (LIBS). The characteristic lines of C, Mg, Al, Ca, Si and Fe were observed in the spectra. Based on the spectral similarity between pencil marks and pencil lead, the spectrum data of pencil lead were used in the data analysis in order to improve the signal-to-noise ratio (SNR). Principal component analysis (PCA) was conducted to build a model to indicate the sample in the principal component space. BP neural network and k-nearest neighbor (KNN) based on PCA were employed to build writing marks identification models, which achieved the identification accuracy of 97.9% and 98.33%, respectively. As revealed by the results, LIBS combined with machine learning can serve as a novel method for document examination, and it has promising applications in the judicial field.