Researchers have developed a computer algorithm that can identify some plant species according to their unique sonar echoes. The experiments were meant to help biologists understand how bats find their favorite fruits or insects, but the research might also help engineers design high-speed systems to identify everything from widgets on conveyor belts to faces in crowds.
Bats might be legally blind, but they can fly straight to a desired fruit tree, even one growing amid dense foliage. They do so using a process called echolocation, in which they send out a series of chirps and then listen very carefully to the returning echoes. Inspired by this ability, researchers in Tübingen, Germany, decided to see if they could invent an artificial system that would perform the same task.
First, the team developed data sets called spectrograms by bouncing sonar signals off five kinds of plants, including spruce trees and black thorn bushes. The researchers then characterized the echo response time and frequency of the resulting sound reflection patterns, which varied according to the number and size of the branches and leaves on each plant. The resulting computer program, says biophysicist and lead researcher Yossi Yovel of the University of Tübingen, could distinguish similar plants with "surprisingly high accuracy." Eventually, the team was able to achieve near-100% success in identifying all five plant species used in the tests, as reported today in PLoS Computational Biology.
The findings will be valuable not only in understanding how bats echolocate, says Yovel, but they should help humans as well. The vast majority of remote-sensing algorithms are based on vision, he says, so if the sonar algorithm can be perfected, one of its advantages will be the ability to function in low light or darkness. (Infrared can't deliver the same degree of resolution.) That could be useful in picking out a crime suspect walking along a dark city street or hiding amid a crowd on a darkened mass-transit platform.
The research could turn out to be "major," says computational biologist Sorin Istrail of Brown University. The idea that a simple algorithm could provide a way to extract a meaningful model for bat echonavigation through tree environments is "remarkable," he says, and could lead the way to practical advances in the machine-learning field. And neuroethologist Steven Phelps of the University of Florida, Gainesville, says the research confirms that subtle differences in the qualities of echoes are enough for a bat to tell a spruce tree from a birch tree. "When we say apples and oranges, we generally assume the differences are obvious," he says, but "I can't imagine having to listen for them."