Mobile app for identifying recyclable items using a convolutional neural network

This project focused on developing a mobile app that can easily identify recyclable items and provide instructions for recycling or disposal. The use of such a device can help in recycling appropriate items, increasing the number of items that a household might recycle and also improve the effectiveness and speed of recycling at the recycling plant. In order to easily identify items, the app uses the phone’s camera to allow a person to easily take a picture of an item. The first phase of the project created a mobile app that sends the photo to a cloud server running a Convolutional Neural Network (CNN) that performs the classification. The server application then looks up recycling instructions on a MySql database and sends back the instructions to the mobile phone app. The mobile app also has a local version of the database so that items can be looked up manually. A key focus of the project explored the public datasets that were available for recycling, training various mobile-oriented CNNs to determine their effectiveness using these public datasets, tuning hyperparameters for optimum performance of each model, and selecting the best CNN, dataset and framework that could ultimately allow the app to be deployed to run natively on a mobile platform.