Watt's That? | Devpost

Watt's That? Logo
Capacitor Detected UI
LED Detected UI
No Component Detected UI

Inspiration

Before women-focused engineering toys like GoldieBlocks emerged on the market, many women who are engineering students today did not grow up with the toys and tools to introduce them to engineering like men. We wanted to create a product so women who are not in electrical engineering fields but still use electronics, like biomedical engineering, can have a quick, easy, and discrete explanation of different devices. This will prevent them from having to ask male students in their class where they risk gaining a condescending explanation. All you have to do is hold up the component to the camera to have a part name and crash course graphic displayed!

What it Does

This app classifies electrical components. For this prototype, we classify LEDs and capacitors and give a rundown of the most essential information so you don't have to spend hours searching on the internet or asking someone else.

How We Built It

This project involved building and training a custom convolutional neural network (CNN) and training it on over 500 images of custom data split between variations of images of capacitors, LEDs, and images with nothing in it!

After training the model, we tested it and used OpenCV to access the laptop's camera. We fed that footage into our CNN and then overlayed a graphic onto the image being displayed depending on what the network's prediction is.

Challenges We Faced

Image classification is really difficult, especially if you don't have access to a powerful GPU. We attempted to remedy this by training on Google Colab, but we ran out of GPU space multiple times and also ran out of drive space because of the number and size of our images. This meant we had to be selective with which weights we were able to save and we also had to decrease the size of our network, making it less powerful.

Another issue we ran into was data. Many of the components were too small on the screen, and when the network performed transformations during the pre-processing portion, the image became too dark to discern which images had which component, if any.

Accomplishments that we're proud of

Not only did we make an easily comprehensible UI for component education, but we built our network from scratch! We learned a lot about tiny parameters that make a huge difference when training a network!