Robustness of texture-based roundwood tracking.

Area of Science:

• Forestry Science
• Computer Vision
• Biometrics

Background:

• Increasing importance of wood log origin verification for Industry 4.0 and combating illegal logging.
• Previous wood log tracing methods lack practical application across diverse processing stages.
• Need for robust tracking solutions from forest to sawmill.

Purpose of the Study:

• To investigate the feasibility of cross-dataset wood log tracing between different processing stages.
• To evaluate various image-based methods for tracking individual logs.
• To assess the impact of different imaging domains (RGB, CT) on tracing accuracy.

Main Methods:

• Utilized image data from 100 logs across multiple stages (forest, lab, sawmill) including CT scans.
• Conducted cross-dataset experiments comparing forest, RGB sawmill, and CT sawmill datasets.
• Employed Convolutional Neural Networks (CNNs), shape descriptors, and biometric recognition techniques (iris, fingerprint).

Main Results:

• Demonstrated the feasibility of wood log tracing between different processing stages and imaging domains (RGB-CT).
• Tracking accuracy is contingent upon the visibility of annual ring patterns or shared woodcut features across datasets.
• CNN-based methods and biometric approaches showed promise in cross-dataset log identification.

Conclusions:

• Cross-dataset wood log tracing is achievable, offering a potential solution for supply chain transparency.
• Image quality and consistency of log cross-section features are critical for successful tracing.
• The study highlights the potential of advanced image analysis and biometric techniques in forestry applications.