Many tasty plants look like their deadly doppelgangers. This can cause real problems, so we created Cognize, an app that uses computer vision to distinguish friend from foe.

What it does

Cognize uses a deep vision AI model trained on the Microsoft Azure platform to distinguish between pairs of easily confused poisonous and edible plant species. It also functions offline making it a great survival tool.

How we built it

-- web app built with JavaScript frontend and python backend -- each image captured by the web app is analyzed by way of a call to the Microsoft Azure prediction API -- to train the Microsoft Azure model we fed it labeled images from google, some of these images were selected by hand and some were gathered by a web scraper that we built --template on GitHub for the Azure training model exported through Tensorflow

Challenges we ran into

-- training the neural network was a big challenge. Each time we added a new tag and set of pictures the precision changed so we had to make adjustments each time to keep the neural network functional.

Accomplishments that we’re proud of

-- proud of a working product in the end (trained neural network, android app)

What we learned

-- how to

What's next

-- adding more species! -- another interesting use case is detecting native vs invasive species in a given geographic location

Built With

Share this project: