AUNet: a deep learning method for spectral information classification to identify inks.

Area of Science:

• Forensic Science
• Analytical Chemistry
• Computer Vision

Background:

• Document tampering and contract forgery are significant legal challenges.
• Identifying different inks is crucial for forensic document examination.
• Existing methods may be destructive or lack precision.

Purpose of the Study:

• To develop a non-destructive method for identifying different inks using hyperspectral imaging.
• To propose an attention-based U-Net (AUNet) model for accurate ink classification.
• To enhance the detection of illegal document tampering and contract forgery.

Main Methods:

• A hyperspectral system was employed to capture spectral information from various inks non-destructively.
• A U-Net architecture was utilized for deep fusion of multi-level spectral data, addressing small sample size limitations.
• Spatial and channel attention modules were integrated into the U-Net to focus on critical features.

Main Results:

• The AUNet model achieved high performance in classifying ink spectral information.
• Specific performance metrics include 97.81% accuracy, 98.71% recall, 98.80% precision, and 98.71% F1-score.
• The proposed method demonstrates effectiveness in distinguishing between different ink types.

Conclusions:

• The AUNet model, combined with hyperspectral imaging, offers a robust solution for forensic ink analysis.
• This approach provides an effective tool for detecting illegal document tampering and forged contracts.
• The study highlights the potential of deep learning with attention mechanisms in forensic science applications.