Segregating Hazardous Waste Using Deep Neural Networks in Real-Time Video
Dorothy Hua, Julia Gao, Roger Mayo, Albert Smedley, Piyush Puranik, Justin Zhan
Abstract
Sustaining a society requires reusing, reducing, and recycling waste. Waste disposal has always been a problem in developing countries because of inadequate infrastructure. By utilizing artificial intelligence to detect hazardous waste, more individuals will be protected from the negative effects of it. To help mitigate this problem, we experimented with Keras, to create a convolutional neural network, and OpenCV, to create real-time videos, that identifies hazardous waste from other recyclable materials. Through the use of machine learning, our model is able to categorize different recyclable materials with about 90% accuracy. Objects within the video receive a prediction for 3 classifications which includes batteries, syringes, and nonhazardous waste. Then, the category with the highest category is what the network will classify it as. In conclusion, the model is able to identify hazardous objects and recyclable items within a pile of trash to help protect all individuals.