Calligraphy and Painting Identification 3D-CNN Model Based on Hyperspectral Image MNF Dimensionality Reduction

Area of Science:

• Art authentication
• Cultural heritage preservation
• Image processing

Background:

• Calligraphy and paintings hold significant cultural and economic value.
• Forgeries threaten the integrity of art markets and cultural heritage.
• Efficient and intelligent identification methods are crucial for art authentication.

Purpose of the Study:

• To develop an efficient, accurate, and intelligent technical identification method for calligraphy and paintings.
• To combine hyperspectral imaging with deep learning for improved art authentication.
• To reduce data redundancy and computational load before deep learning analysis.

Main Methods:

• Utilized hyperspectral imaging technology for material attribute recognition and imaging detection.
• Applied Minimum Noise Fraction (MNF) dimensionality reduction to compress hyperspectral data.
• Developed a deep learning model using 2D-CNN and 3D-CNN architectures with a core structure of 4 convolution, 4 pooling, and 2 fully connected layers.

Main Results:

• Both 2D-CNN and 3D-CNN models demonstrated high accuracy in identifying calligraphy and paintings.
• The 3D-CNN model exhibited superior learning convergence and stability compared to the 2D-CNN model.
• The 3D-CNN model achieved identification accuracies of 93.2% for author and 95.2% for authenticity on the test set.

Conclusions:

• Deep learning methods, particularly 3D-CNN combined with MNF dimensionality reduction, offer a powerful approach for authenticating calligraphy and paintings.
• This technique enhances the efficiency and accuracy of art identification, aiding in the protection and fair trade of cultural art.
• The study highlights the potential of hyperspectral imaging and advanced deep learning for preserving cultural heritage.