Corally | Devpost

Thumbnail
Image Area
Personal Information
Data Flow Diagram

Inspiration

I've seen crazy headlines on the trajectory of coral reefs and it always blew me away to see how negative it was. In the past, I've researched the impact of coral reefs on the ocean ecosystem and how important it is that they stay healthy. Recently, I've also been furthering my knowledge in machine learning and decided to combine the two things.

What it does

Corally is a web app that ensures that coral reef organizations always have up-to-date information on the status of coral reefs with the help of users across the world. When a user opens Corally, they are prompted to upload a picture of coral that they have taken. The image is then processed and passed through a custom deep convolutional neural network. The model predicts whether the coral is bleached, dead, or healthy; these predictions are then outputted to the user. After, the user enters details about the location of where the picture was taken, such as the country, city, beach name, etc. They go on and select the date that the picture was taken on and enter in personal information in case the receiving coral reef organizations would like to contact them with questions about their submission. The about page has information about things like Corally’s mission and future plans for Corally. It also covers how Corally works with a diagram.

How I built it

Corally's web app was built using a framework called Streamlit, which is used to create and host web apps in Python. The model for determining the health of coral images was created using TensorFlow and Keras. This model is an 11-layer convolutional neural network with 9 hidden layers and 3 convolutional layers. It was trained on a dataset of dead, bleached, and healthy coral for 20 epochs.

Challenges I ran into

The main challenge I ran into was the creation of my model. It was hard to convert uploading images into a supported data type. I also had trouble with the accuracy of the model; the model has trouble identifying bleached versus dead coral, and also seems to lean towards healthy coral. Sharing by project with Streamlit also gave me some issues.

Accomplishments that I'm proud of

I'm proud of the fact that I uploaded a working prototype for the first time. I'm also proud of the model I created.

What I learned

I learned how to create a convolutional neural network with TensorFlow and how to upload a website with Streamlit.

What's next for Corally

In the future, I hope to further optimize the TensorFlow model as it struggles with bleached versus dead coral images. It also has a strong bias towards healthy coral. I also plan to partner with as many coral reef organizations as possible in order to maximize the impact of Corally.