Microsoft Project Uses Machine Learning To Identify Flowers

You can use your phone to figure out what song is playing on the radio, and there are even some apps that try to figure out how many calories are in your meal based on a snapshot of it. Now, thanks to Microsoft Research Asia and the Institute of Botany Chinese Academy of Sciences, there's a way to do this with flowers. The idea is that we're surrounded by flowers all summer long, but probably can't identify many of them. If a particularly striking one finds your eye, you can use this recognition project to upload a photo of the flower and get information about it in return.

Microsoft calls the new project the Smart Flower Recognition System, and it was born from an idea about pairing Microsoft Research's image-recognition technology with a massive database of flower images. Because there are about a quarter of a million flower species in existence, such a database would be a godsend to botanists who otherwise may struggle identifying the exact type of flower they encounter.

The image database contained about 2.6 million photos of flowers that were uploaded by the public. The researchers first set out to teach the smart system to recognize images and sort away the 'bad' ones. The end result was taking about 800,000 of those photos and pairing them with the deep-learning network Caffe.

Over time, researchers got the smart recognition system nearly on par with humans — it can recognize what type of flower it sees with a 90+ percent accuracy rate. Soon enough, says Microsoft, devs will make applications based on this smart flower recognition system, some of which, it seems, will be targeted at the everyday public as well as field experts.

SOURCE: Microsoft Research