Research on Ink Speed Recognition Method of Hyperspectral Imaging Ink Pad Based on Machine Learning

In this paper, the classification models in machine learning were combined with hyperspectral imaging technology to identify different brands or different models of stamp-pad inks of the same brand. Spectral acquisition of stamp-pad ink was performed under a hyperspectral imager, and the acquired spectral data was combined with machine learning classification algorithms to build an identification model and predict the class of the delineated prediction samples. The algorithms employed include linear discriminant analysis (LDA), parameter selection of support vector machines based on grid search (grid-SVM) and parameter selection of support vector machines based on particle swarm optimization algorithm (PSO-SVM). After evaluating the identification models established by LDA, grid-SVM and PSO-SVM, the results show that PSO-SVM realizes the best classification results, with the accuracy for six different classes of stamp-pad inks of 96%, with further accuracy for a total of 10 various labels under each class of 95%, thereby achieving the discrimination of ink types that cannot be distinguished by conventional optical methods. Therefore, the model for the identification of stamp-pad ink types that are inaccessible by conventional optical methods was established. Ultimately, the identification of stamp-pad ink types can be realized by combing machine learning with hyperspectral imaging technology, providing a new, scientific and non-destructive examination method for seal examination, document addition and alteration, etc. It also provides an opportunity to build a new platform for the forensic science ink examination.