LeafMachine

LeafMachine extracts phenotypic leaf trait data from digitized herbarium specimens and leaf images using machine learning for automated measurement of leaf morphology to support evolutionary and ecological research.

Key Features:

• Machine Learning Algorithms: LeafMachine employs Convolutional Neural Networks (CNNs), Support Vector Machines (SVMs), and Computer Vision (CV) algorithms as an ensemble to identify and measure leaf traits from images.
• Training and Performance: The models were trained on 2685 specimens from 138 herbaria and extracted leaf measurements from 82.0% of high-resolution images and 60.8% of low-resolution images.
• Error Analysis: Among specimens where measurements were not extracted, 0.9% of high-resolution and 2.1% of low-resolution images were visually determined to have measurable leaves.

Scientific Applications:

• Large-scale phenotypic data extraction: Enables increased access to leaf trait measurements from herbarium collections to support studies in plant evolution and ecology.

Methodology:

Models were trained on a diverse dataset of 2685 herbarium specimens from 138 herbaria and integrate CNNs, SVMs, and CV algorithms for feature extraction and measurement.