There are at least 250,000 species of flowers on our planet. The most passionate gardener or even the most experienced botanist couldn't name all of them. I'm having trouble just learning the names of all of the wildflowers in Texas and that's a very small subset.
Recently, a group of botanists from Institute of Botany, Chinese Academy of Sciences (IBCAS) were having trouble getting consistent results from their crowd-sourced field studies of regional flower distribution. The issue was that the researchers needed more accurate flower identification methods. The team happened to hear Yong Rui from Microsoft Research Asia speak at a conference about their new facial recognition technology and they realized that if software could be created to learn faces, it could also be made to learn flowers.
The team met with Rui and soon the two parties realized they could help each other -- the botanists could have a way to more accurately identify flowers for their studies and Microsoft would have an interesting initiative to help improve its software -- and the Smart Flower Recognition System was born.
The botanists already had a collection of 2.6 million images of flowers that had been uploaded and identified by members of the public. One problem was that many were unclear or bad pictures that needed to be skipped over. The Microsoft team first created an algorithm to filter out the bad photos and then started the more complicated task of creating a tool that could recognize tiny differences between species of flowers in order to identify them.
The team trained a deep neural network to recognize flower species using a set of learnable filters that looked for specific details at certain points across the width and height of the flowers.
The researchers then ran million of images through the system and it was able to correctly identify the flowers in the photos with around 90 percent accuracy, which while just shy of what humans can do, it is much faster. The botanists were able to use the system to meet their research goals and now the researchers want to develop tools that anyone, botanists or just the curious, can use.
“The flower-recognition engine enables domain experts to acquire plant distribution in China in an efficient way," said Zheping Xu, assistant director of IBCAS. "Not only that, this engine can help ordinary people who have a strong interest in flowers to gain more knowledge."
Microsoft and IBCAS hope to develop an app for your smartphone soon.