Deep learning algorithms were trained to recognize and classify thousands of plant and animal specimen, with great results.
Can you tell the difference between a cheetah and a leopard, an alpaca and a llama, a donkey and a mule?
What about plants? Can you differentiate a daisy from a chamomile flower? A rose from a begonia?
No? Neither can I! But, for my part, I decided a long time ago I’d let computers do all of my thinking.
Finally, they’re catching me up.
To be fair, those were common examples of animals and plants that are easily confused. There are several millions of species that share many traits while being far apart–biologically speaking.Deep learning recognizes and classifies thousands of plant and animal specimen.Click To Tweet
If you have a plant whose name you don’t know, you now have the option to upload photos to plant identification groups on Facebook and let members identify it for you.
You can also use a specialized app that allows you to identify a plant based on one or more photos, which is effective in some cases, but not always.
Deep Neuronal Networks Trained to ID Animal and Plant Species
A few years ago, deep learning networks were having a hard time telling a cat from a human, but now, they’re able to identify thousands of species.
In 2012, Google’s neural network made headlines after it taught itself to differentiate between the shapes of humans and cats with 70% accuracy.
Now, merely five years later, a contest asked researchers to train an AI to ID over 5000 different species of plants and animals.
Under the aegis of Google, the contest, called the iNaturalist Challenge 2017, was held between June and July on Kaggle by Grant Van Horn, a graduate student at the California Institute of Technology.
As part of the iNat dataset, there were 675,000 training and validation images, corresponding to 5,089 plant and animal species. This was put at the disposal of the participating teams.
Final results of the iNat challenge are yet to be disclosed, but some teams claim their neuronal networks managed to get 85% prediction accuracy.
Taxonomic AI: Automatically Classifying Thousands of Herbarium Specimen
A herbarium is a collection of dried and pressed plants, put between sheets of paper, usually carefully composed by researchers.
With an estimated 350 million plant species in the world, only a small portion of that has been digitized.
Erick Mata-Montero of the Institute of Technology-Costa Rica and botanist Pierre Bonnet of the French Agricultural Research Centre for International Development have contributed to the Pl@ntNet app project by doing the digitizing work.
Pl@ntNet is an app that helps to identify plant specimen based on pictures.
Now, as reported by Nature, the team trained a similar AI algorithm to identify thousands of dried plants composing a very rich herbarium.
The research duo fed their algorithm with 260,000 pictures taken from herbarium sheets, corresponding to over 1,000 different species.
The AI algorithm took up the challenge and made a correct answer 90% of the time intially. Then, it finished with an 80% accuracy (correctly identifying 208,000 plants)–enough to make a human taxonomist green with envy.
Both initiatives demonstrate the significant AI progress made in recent years and speak volumes about the advances ahead and the innumerable opportunities that are opening up with this technology.