Fossilised leaves are forced to tell their story
Millions of fossils around the world could provide insights into plant evolution, but until now they have been too hard to identify. Viviane Richter reports on how computers are changing this.
Plant fossils gathering dust in museums around the world may have more stories to tell yet. An international team of researchers has developed a computer vision program that identifies a plant's family from a single leaf image. This program, which they hope to use on fossils, may provide scientists with a better tool to study plant evolution.
To categorise a new plant specimen to a family tree, botanists rely on the fruits, seeds and flowers – even though plant leaves are far more common.
The problem is complexity, says study author Peter Wilf from Penn State.
Where leaf patterns have been described in detail since the 1800s, the descriptions are often not enough to distinguish between different plant families. That limitation has left palaeobotanists scratching their heads over many unidentified fossil leaf samples.
Wilf’s team suspected computers may be able to pick up more points of difference than the human eye.
His team bleached and stained more than 7,500 modern leaf samples to reveal the pattern of veins in each, producing a large collection of images that resembled fossil imprints.
Meanwhile, the team modified a computer vision program, which had originally been used to spot animals in photographs, by kitting it out with a codebook of leaf shapes and venation patterns.
The program learned to classify each image into plant families and orders, by generating “heat maps” which pinpoint and rate different leaf signatures.
Using this program, the team accurately assigned 72% of leaves to more than 19 different families. The program also categorised leaves of the coffee plant – one of the hardest families to identify, the authors say – to 90% accuracy.
“It normally takes a trained person a few hours to describe one leaf according to the standard protocol,” said Wilf. “The computer program is thousands of times faster, automatically generates a dictionary of more than 1,000 elements and then actually shows us what parts of the leaf are diagnostic.”
The goal of the research is to apply the program to the millions of plant fossils in museums around the world, to see if it can allow scientists a more accurate picture of plant evolution.
Leaf fossils are one of the most under-used resources for understanding plant evolution, Wilf says.
“Variation in leaf shape and venation, whether living or fossil, is far too complex for conventional botanical terminology to capture. Computers, on the other hand, have no such limitation.”