A computer-vision method has enabled an international team of researchers to quickly classify leaves and generate vast quantities of new botanical data.
The system is not programmed to exhibit a particular behavior, but instead it learns from leaf images together with category labels corresponding to family and order.
The researchers say they have achieved a 72-percent accuracy rate over 19 leaf families compared to about 5 percent for random chance.
The researchers provide the program with half the images already identified so it can automatically learn a dictionary of special features, which are critical to identifying leaves. The system also learns to ignore the typical problems of low image quality, insect bites, and mounting defects, and the algorithm then receives unlabeled test photos and uses its dictionary to identify them. The researchers repeated this procedure 10 times, randomly selecting the training and test images; the outcomes agreed, with only a 1-percent difference between runs.
The computer generates a "heat" map directly on the leaf image, identifying and rating areas of importance for correct identification.
"Variation in leaf shape and venation, whether living or fossil, is far too complex for conventional botanical terminology to capture," says Pennsylvania State University professor Peter Wilf. "Computers, on the other hand, have no such limitation."
From Penn State News
View Full Article
Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA