This project focuses on real-time waste classification using Convolutional Neural Networks (CNNs) deployed on a Raspberry Pi 4B. The goal is to classify waste objects efficiently and accurately, making it suitable for edge computing applications.
The project utilizes the Recycled dataset created by Portland State University. This dataset provides a diverse collection of waste objects suitable for training and testing various machine learning models. Initially, the models are trained and evaluated using this dataset to establish a baseline performance.
The Roboflow Recyclable Items dataset supplements the training data, offering additional variety and depth to enhance model performance further.
The notebook compares various machine learning models, including CNN architectures, for waste classification. Initially, these models are trained and evaluated using the Portland State University Recycled Dataset. The best-performing model, which was determined to be a CNN, is further refined and compared with other CNN architectures using the Roboflow dataset.
Once the best CNN architecture is identified, it is deployed on the Raspberry Pi for real-time waste classification.