What exactly is that flower over there? The Institute of Botany at the Chinese Academy of Sciences is teaming up with Microsoft Research to solve the undying question many people have in mind. With a project called the Smart Flower Recognition System, machine learning and recognition is being incorporated into an application that will make identifying flowers much easier with just a picture.
After listening to Yong Rui’s seminar explaining image-recognition technology and its capabilities, the IBCAS botanists were inspired. In collaboration with Microsoft Research Asia, they began developing an engine that could harness these technologies towards recognizing plant related images. Unfortunately, nearly 2.6 million images were quickly submitted to data storage without any filtration to speak of. Jianlong Fu provided a solution through trial and error, building tools that filtered out bad images and discerned the intricacies of each species of flowers.
The Smart Flower Recognition System was then integrated with a 20-layer neural network, sifting through the thousands of images submitted. With nearly 800,000 pictures, the engine accurately identified more than 90% of flowers. The achievement is close to the success rate of human botanists.
The recognition software has more capability than just recognizing flowers. The Caffe deep-learning framework developed by the Berkely Vision and Learning Center focused on autonomous perception research. It’s already used for academic research projects, prototypes, and industrial applications of a large variety. Through the use of the neural network library, developers and users can find tags for images in a minimal amount of time. Interested parties looking into the framework’s utility can experience a minor demo on the Berkeley Vision site by providing an image URL or uploading a file.
More technology is evolving every day towards progressive machine learning and recognition. For now, Chinese botanists are grateful for the tool and hope to see developers build applications for real world application to sate the curious flower hunter.