SORTR

The Project

SORTS is a Machine Learning based interactive device to help sort waste at the time of disposal. This would provide a preemptive, budget-friendly solution that can be deployed at scale. Model is trained using Teachable Machine, a web-based tool from Google, for creating custom machine learning models. System Prototype is implemented on a Raspberry Pi 4 computer. Object classified using ML model after scanning with camera. Uses color-coded LEDs, and an LCD screen for disposal bin identification. Audio clips played on speaker to provide educational facts about disposed items.

View this project on YouTube

Team Comments

Varun’s science teacher runs a study hall event where students properly sort trash. EPA estimates increasing our national recycling from 30% to 35% would reduce greenhouse gas emissions by 10 million metric tons of carbon. Both of these got Varun thinking that improper sorting is a big problem.

Training custom models on Teachable Machine. Too much object data was causing errors to creep in, reducing efficiency. Increasing the epoch value(number of iterations for each scan) helped. Training under better lighting conditions helped.

Have the model based detection happen in a more capable machine(Cloud servers). The local device could send the captured image and get the response back. Train with more waste items to make it more holistic.