Recyclify

Inspiration

The spark for our project ignited during a routine moment—disposing of lunch trash with a friend. When my friend nonchalantly tossed an aluminum can into the trash bin, I was struck by his surprise at my suggestion to recycle it. This exchange unveiled a widespread unawareness about recyclables. Determined to bridge this knowledge gap, I envisioned a simple tool to demystify recycling for everyone.

What It Does

Our solution is a binary image classifier that categorizes items into 'trash' or 'recycling.' Users upload a picture, and our system advises on its recyclability, thereby encouraging responsible waste disposal.

How We Built It

We harnessed TensorFlow and crafted a Convolutional Neural Network (CNN), training it on a Kaggle dataset originally featuring 12 categories. We manually reclassified these into our two target categories, tailoring the dataset to our needs.

Challenges We Faced

Our foray into transfer learning with pre-trained models like VGG-16 and ResNet50 initially fell short of expectations. Pivoting to a custom-built CNN, we achieved a training accuracy of 90% and a test accuracy of 87%. However, real-world testing hovered around 64% accuracy. Limited by the processing capabilities of our laptops, training and testing were painstakingly slow, often taking 30 minutes to an hour for a single iteration, which considerably decelerated our progress. Because of this, we couldn't train it on larger and more robust datasets to make the model well-rounded.

Accomplishments We're Proud Of

Despite our limited prior knowledge, we managed to develop a functional neural network from the ground up—something we are extremely proud of.

What We Learned

Our journey was rich in learning, particularly about the intricacies of CNNs and the various metrics and tests to evaluate a neural network's performance.

What's Next for Recyclify

Looking ahead, we aim to evolve Recyclify from a web-based interface to a mobile app incorporating in app object detection and classification using the phone's camera for a more interactive experience. We also plan to integrate a chatbot to answer recycling queries and establish a database for storing scanned images. This will not only refine our model with real-world data but also offer valuable insights into waste management trends.

Built With

• python
• streamlit
• tensorflow