Nondestructive and Rapid Identification of Stamp Pad Ink Based on Hyperspectral Imaging and Extreme Learning Machine
Abstract
The examination of stamp pad ink in questioned documents serves as a crucial scientific basis for forensic authentication. This study presents a novel rapid classification framework integrating Hyperspectral Imaging (HSI) and Extreme Learning Machine (ELM) to address the challenges of timeliness and accuracy in nondestructive ink detection. A total of 24 photosensitive ink samples from 21 brands were collected, generating 72 standardized stamped impressions. Spectral-spatial data were acquired using an HSI system (400-1000 nm, 5 nm spectral resolution). preprocessed by Multiplicative Scatter Correction (MSC) to mitigate substrate interference. Experimental results demonstrate that the HSI-MSC-ELM framework achieved an accuracy of 98.38% on the test set without feature dimensionality reduction (full 121 spectral bands), outperforming Random Forest (RF) by 4.63% and Backpropagation Neural Network (BPNN) by 6.34%. Crucially, the detection time was only 1.59 seconds – 28× faster than RF (45.90 s) and 285× faster than BPNN (453.36 s). This approach provides a simple, nondestructive, and efficient solution for forensic document examination, with potential to replace traditional destructive techniques.