If you find you can never quite remember the names of all the wildflowers growing along your cottage road – or for that matter, what’s growing in your backyard garden – they may soon be an app for that, thanks to work done by Microsoft Research Asia.
The Smart Flower Recognition System promises to identify a flower from the over 250,000 species in existence with 90 per cent accuracy, according to a blog post from Microsoft. The technology arose from a partnership between Microsoft Research Asia and the Institute of Botany, Chinese Academy of Sciences. After meeting at a conference, the institute sought Microsoft’s help with categorizing its database of 2.6 million images of flowers. The image set was also growing rapidly, since it accepted uploads from around the world.
To solve the problem, Microsoft’s team created algorithms that first weeded out the bad images of flowers. From there, it was on to training a “20-layer convolutional neural network” to scan every photo pixel by pixel, creating a two-dimensional map. After scanning enough images, the network learns the common qualities shared by images of the same flowers.
After processing 800,000 images from the database, Microsoft was able to correctly identify flower images more than 90 per cent of the time.
While the image recognition engine is only accessible to the institute for now, Microsoft expects that soon enough mobile developers will be able to tap it to create apps for the consumer market.