Computer vision cracks the leaf code.

Area of Science:

• Botany
• Computer Science
• Evolutionary Biology

Background:

• Angiosperm leaf shape and venation are complex and variable, posing challenges for botanical classification.
• Previous computer vision efforts focused on species-level identification, leaving higher taxonomic ranks unexplored.
• Classifying major evolutionary groups like families and orders requires understanding extensive foliar variation.

Purpose of the Study:

• To test if computer vision algorithms can learn and classify leaf features at the family and order levels.
• To explore the potential of machine learning in analyzing large botanical specimen datasets.
• To identify novel morphological characters with phylogenetic significance.

Main Methods:

• A database of 7,597 cleared leaf images from 2,001 genera was utilized.
• Machine learning algorithms created a visual codebook of leaf shape and venation patterns.
• A computer vision system was trained to classify leaf images into botanical families and orders.

Main Results:

• The automated system achieved a classification success rate significantly greater than chance for families and orders.
• Diagnostic features were visualized using heat maps, aiding in the recognition of novel morphological characters.
• The study demonstrated the feasibility of using computer vision for higher-level plant classification.

Conclusions:

• Computer vision and machine learning are powerful tools for analyzing complex botanical data.
• This approach can significantly advance systematic and paleobotanical studies by uncovering new morphological insights.
• Leaves, analyzed via computer vision, are poised to contribute substantially to understanding plant evolution.