Bloom - Hanging Basket Flowers Identification App

• 

  Bloom Demo

Inspiration

During the longest lock-down in Ontario, the demand for gardening has been increasing. More and more people consider flowers and plants as their “friends and families” due to the long hours they have been spending with these lovely creatures. Our group was inspired by the popular view of flowering plants that grow in hanging baskets on people’s balconies, enriching the colour of a summer day. We wanted to create an app that would be able to identify hanging basket flower species in Ontario so that people would get an idea of the name of the species they are interested in and they would be able to easily purchase them through online flower markets.

What it does

Our app aims to help users identify the 15 most common hanging basket flower species in Ontario. When users find a hanging basket flower species that they are fond of, they can first take a picture of it. Then, download our code from our GitHub link and run the commands. After that, a window would pop up and ask them to insert the picture for identification. Users can upload the picture from a local source, and an identifying result would show up beneath the window. The result would return either an identification that includes how much percentage of the picture is similar to a species, or a negative identification, which states that the species was not included in the app, which means the plant is not grown in Ontario.

How we built it

We use Keras to train an Xception network for the training part, and we use PyQt5 to create GUI for fitting.

Challenges we ran into

We encountered the problem of having too many irrelevant photos in our training set because all training photos are downloaded at once. Since a large number of photos were downloaded, it was difficult to manually delete all the irrelevant photos.

Accomplishments that we're proud of

We were able to identify 15 species with high accuracy with 60% accuracy for several hours of training. The accuracy will go up for a longer training time.

What we learned

Technically, we learned how to use Kares and how to use PyQt5. From the perspective of gardening, we were able to learn about the details of each of the 15 hanging basket flowers, including colour, bloom season, light requirement and companion plants.

What's next for Bloom - Hanging Basket Flowers Identification App

We encountered the problem of having too many irrelevant photos in our training set since all training photos were downloaded with all relevant searches on the internet. We were hoping that we would be able to expand the species that were included in our identification app, such as adding other classifications of gardening plants in Ontario. For example, cut flowers for display or house plants. We were also hoping to increase the accuracy of our app by entering more precise training photo sets in it. If time permits, we would be able to hand-select the photos of species that go into the training set for better identification purposes. Lastly, we were hoping to expand the training trials to a larger number, if time permits.

Built With

• keras
• python
• tensorflow