Understanding and classifying different environmental factors will be an important aspect in large scale remote sensing and climate progressive technology. While deploying machine learning algorithms on a local machine is supported by various frameworks, it is difficult to intergrate these techniques on a real-time website. The goal of this project was to access a open source dataset of satelliete imagery, train a CNN, and then deploy this model on a website that allows for user input into predictions.
While our website is simple, our team is excited because we feel this is the foundation to more robust and accessible climate modeling.
We used replit, flask, and python for the web application side. We used Tensor Flow, Keras with a VGG16 CNN architecture for the machine learning aspect. We trained our model in the google cloud with an acceleralated GPU compute, which required a difficult configuration. Our code allows for us to easily access the cloud API and supports a machine learning model that can give predictions without signficant latency.
Our web app takes in a satellite image and uses a machine learning model to predict which type of biome the image is a part of. We have almost no experience building anything like this (or anything at all honestly, seeing how this is our first hackathon, and three of us are first years) so everything we did was a huge challenge. Overcoming those challenges taught us so much. We want to add more functionality and potentially publish it to the World Wide Web in the future.
Log in or sign up for Devpost to join the conversation.