Accurate classification of fresh and charred grape seeds to the varietal level, using machine learning based...

Area of Science:

• Archaeobotany
• Genetics
• Taxonomy
• Computational Biology

Background:

• Grapevine (Vitis vinifera L.) has thousands of cultivars, historically identified by ampelography, now mainly by genetic analysis.
• Genetic analysis is often unsuccessful for archaeobotanical remains due to poor genomic preservation.
• Morphological analysis of grape pips has been attempted using 2D imaging, but with limited success.

Purpose of the Study:

• To develop a highly accurate varietal classification tool for grapevine seeds.
• To overcome the limitations of genetic analysis for identifying ancient grape remains.
• To establish an accessible and effective method for grape seed classification.

Main Methods:

• A novel 3D seed scanning approach was developed.
• Machine learning algorithms, including Iterative Closest Point (ICP) registration and Linear Discriminant Analysis (LDA), were applied.
• The methodology was tested on both fresh and charred grape seeds.

Main Results:

• The 3D scanning and machine learning approach achieved 91% to 93% average classification accuracy.
• The method demonstrated effectiveness in classifying fresh and charred seeds.
• A "tournament" approach further enhanced accuracy when classifying eight distinct groups.

Conclusions:

• The developed 3D seed scanning methodology offers a highly accurate and accessible tool for grapevine varietal classification.
• This approach significantly improves the identification of archaeobotanical grape remains where genetic methods fail.
• Future developments hold promise for advancing archaeobotany and general taxonomy through improved seed classification.