Inkjet ink classification and source prediction based on direct analysis in real-time mass spectrometry (DART-MS) via mass imaging and convolutional neural network (CNN)

Abstract

This paper presents mass spectrometry imaging (MSI) as an intuitive method of showing the differences between inks, and provides a novel method of ink source prediction based on direct analysis in real-time mass spectrometry (DART-MS) spectral analysis. The ink database comprised 106 types of ink from three brands (Canon, Epson, and HP) with high market share. High-dimensionality reduction methods, such as principal component analysis (PCA), non-negative matrix factorization (NMF), and probabilistic latent semantic analysis (pLSA), were used to reduce the dimensions of the mass spectral data. Characteristics and statistical descriptions of the mass spectra of each ink are given in this paper. Then, MSI was used to visualize the results after PCA, NMF, and pLSA. Finally, a convolutional neural network (CNN) was applied to the data after high-dimensionality reductions, which helped predict the ink source. Results showed efficient and excellent performance of machine-learning analysis in ink source prediction using one pixel in MSI (100% for black, magenta, and yellow inks; 99.6% for cyan ink). In addition, a blind test was used to validate the performance of the CNN model, with results showing that the more pixels used, the more accurate the results. Therefore, we strongly recommend wide-area detection of ink to ensure accurate results.