virtualEagle | Devpost

Our first successful Keras model
Home Page
Image classification web app
Image classification web app response

Inspiration

Drastic increase in recycled items over the next few years.
Great Pacific Garbage Patch is only getting larger over the years
Estimated 100 million tons of trash
Human Labor is not effective ## What it does Differentiates recyclable and organic materials using an image classifier based on a Keras neural network. There is one dense layer and 3 pooling layers and in the end, there is a sigmoid with one node. The sigmoid helps show the percentage. ## How we built it We used over 23 000 different images for classifying between organic and recyclable items. We trained a neural network using a cloud VM which we got from Digital Ocean. We designed the model ourselves and fitted the data from Kaggle. ## Challenges we ran into
Overfitting - I forgot to dropout which caused the data to basically memorize the pixels and not be generalized which led to a poor validation score.
Combining Nodejs with Python - We created an Express Server and ran into the issue of sending standard buffer data between two different languages. Python and Javascript. ## Accomplishments that we're proud of
Creating an effective neural network that we _ made ourselves _.
Created a beautiful UI design for the web app
Linked the raspberry pi to our cloud VM ## What we learned
How to debug and properly increase the validation score of neural networks
How to pass data between two different languages
How to effectively build an API for an IoT device (Raspberry pi) to use for image classification project ## What's next for virtualEagle
We would like to train more data and have a better CPU/GPU to train our complex models.
Create a public API for anyone to access through an API key
Cooperate with industries such as Waste Management Inc., and provide modern solutions to their traditional disposal systems.