Inspiration
What it does
It navigates a given scene and notifies whether it has identified a fire. It is 80% accurate on my test data. It functions well on simple scenes.
How I built it
I built a data set with labels "fire and "nofire". I split images into smaller ones to train on the limited context required by the hackathon. I trained a convolutional neural network on the data set. Played with number of layers and amount of data processed to achieve current result. Used Python and Microsoft Azure Machine Learning Services.
Challenges I ran into
Many images of fire I found were very low resolution. I don't have a lot of experience with neural networks so this was a learning experience for me but I greatly enjoyed it. I've never made a data set before.
Accomplishments that I'm proud of
I'm proud that I made my own data set to train on, I had never done this before and was pleasantly surprised by the results.
What I learned
I realized that much of my data focused too much on closer fire and I should've tried to get more images in which fire appears farther away. I think I could improve upon the model if I made a label for smoke and worked towards better differentiating bright smoke from fire.
What's next for Do Drones Dream of Electric Fire?
I may try to improve my data set and play with the model more. Will definitely play with image classification more, I would like to improve on this project after leaning more.
Log in or sign up for Devpost to join the conversation.