Many times we might be on a stroll through a beautiful green garden, and you notice a flower. You may remember the flower, but you forgot the name. Without an idea of the name of the flower, you will not be able to google the name of the flower. For the flower enthusiasts, it can be a brain racking session to find the name. What if you could quickly click a picture of the flower and load it into an application to get the name of the flower?
Smart Flower Recognition System
This has been made possible with Smart Flower Recognition system, which is a new project undertaking by the Institute of Botany at Chinese Academy of Sciences. It has joined hands with Microsoft Research to come up with smart answers to the usual question of flower recognition. The system brings together a rare combination of machine learning with the identification of the flower through a source as simple as a picture. There are over 2,50,000 species of flowers and it is a tough task even for the highly experienced botanists.
How it Works?
The system is rather simple whereby a person who sees a flower wants to know its name can simply click a picture of the flower and sent it to the Smart Flower Recognition System. The machine learning will be able to provide the most accurate answer. This depiction and explanation of the system by Yong Rui at the seminar had impressed the IBCAS scientists who led them to join hand with Microsoft to come up with a system to identify plant images. But, to a greater dismay, the system got overly fed with a good load of over 2 and a half million pictures into the system. There was a need to filter the images, and this was done by Jianlong Fu who provided a screen to send only the right images that provided information about the intricate flower structure.
Any artificial intelligence requires integration with a neural system. The system contained more than 800,000 images, and after the Smart Flower Recognition System was merged with a neural system, the system recognised a good deal of the flowers. The neural system used was an intricate one with 20 layers that went through all the images. The result thus obtained was comparative to that of human scientists.
Caffe from BVLC
The job that was done by the recognition does not confine to flower recognition alone. There are wider impending applications of this recognition software developed through machine learning technology and neural networks. Berkeley is a Computer Vision Group from the University of California that is conducting researches on recognition, reorganisation, reconstruction and other data analysis sectors. Caffe is a learning structure thus created by the Berkeley Vision and Learning Center along with the other members. Yangqing Jia did the project during his PhD.
The project has already achieved full recognition with use in research work, to create other recognition prototypes and even for industrial uses. A neural network library of several million images helps the user to find image tags by comparison with the least time frame. The system has been made available on a minor framework structure through the Caffe information page at Berkeley Vision Site. The user can upload an image or provide the link to get a result of tags for the particular picture. The result is not very clear as the work is in progress to create precise labels. But standard images like animals etc. are recognized by the system.
Future Expectation
Each day is opening up newer areas for development in the field of machine learning and artificial intelligence. But in the current state of things, the Chinese botanists are more than elated to have a flower recognition system. They look forward to a developer coming up with an interactive and efficient application to grab the attention and surprise the flower hunters across the globe.
No comments:
Post a Comment