Our Story & Mission
Born from the
Earth's Need
REGAIA is an AI-powered waste classification system that bridges the gap between human behaviour and responsible recycling — making correct disposal as simple as taking a photo.
What We Are
A Neural Eye for the Planet
REGAIA is an intelligent waste classification web application that uses deep learning to identify the material type of any waste item from a photograph. A user simply uploads an image — or takes one directly in the browser — and the system returns an instant classification along with a confidence score.
Behind the scenes, a fine-tuned EfficientNet-B2 convolutional neural network analyses the visual features of the image against six material categories: cardboard, glass, metal, paper, plastic, and trash. The model is further paired with YOLOv8 for real-time object detection, enabling bounding-box-level identification of multiple items in a single frame.
The result is a tool that demystifies recycling — giving anyone, anywhere, an authoritative second opinion on how their waste should be sorted.
The Name
Why REGAIA?
Names carry intent. We chose REGAIA because it encapsulates the entire philosophy of the project in a single word — one that speaks of renewal, responsibility, and reverence for the living world.
REGAIA
Trash Smarter, Live Better
RE —
From Latin re-
meaning "again", "anew",
or "back to life"
GAIA —
The primordial Greek
goddess of Earth itself —
the living, breathing world
Together —
A declaration that the
Earth can be reborn
through conscious action
In Greek mythology, Gaia is not merely the ground beneath our feet — she is the ancestral mother of all life, the planet as a conscious and sacred entity. By prefixing her name with re-, REGAIA expresses the conviction that the Earth's vitality can be restored — not through grand gestures alone, but through millions of small, informed decisions made every day. Decisions like sorting a bottle correctly. Identifying a can before it ends up in landfill.
Every classification made through REGAIA is, in its own small way, an act of giving Gaia back what belongs to her.
The Problem
What We're Up Against
2.01B
Tonnes of municipal solid waste generated globally each year
~16%
Of plastic waste that is formally recycled worldwide
> 40%
Of recyclable material that ends up in landfill due to mis-sorting
6×
More CO₂ released from waste incineration vs. recycling the same material
The global waste crisis is not purely a production problem — it is fundamentally an identification problem. A significant proportion of recyclable material is contaminated or mis-sorted not out of indifference, but out of genuine uncertainty. Recycling guidelines vary by region, materials look deceptively similar, and the everyday person rarely has access to a reliable reference.
REGAIA attacks this problem at its root: if people can instantly verify what category their waste belongs to, the barrier to correct behaviour collapses. No lookup table, no confusing symbols, no guesswork — just a photo and an answer.
Aims & Objectives
What We Set Out to Achieve
Instant Classification
Deliver accurate waste material identification from a single image in under two seconds, accessible directly in the browser without any app installation.
High Accuracy
Train and fine-tune a state-of-the-art vision model — EfficientNet-B2 — to distinguish reliably between six distinct waste material classes with strong precision and recall.
Universal Access
Ensure the tool works equally well on mobile phones and desktop browsers — including a native in-browser camera feature so no separate device or app is required.
Behavioural Change
Reduce contamination in recycling streams by empowering individuals to make informed disposal decisions — turning passive awareness into active, repeatable behaviour.
Robust Uncertainty
Implement entropy-based confidence gating so the model gracefully declines to classify when presented with ambiguous or out-of-distribution inputs, rather than returning a false answer.
Extensible Platform
Build a modular, well-structured codebase — separating data pipeline, training engine, and inference — so that the system can be expanded to new categories, datasets, or deployment targets.
Tools & Frameworks
The Stack Beneath the Surface
REGAIA is built on a carefully chosen combination of industry-standard tools across model training, object detection, backend serving, and frontend delivery. Every component was selected for its reliability, community support, and suitability to our specific constraints.
Core AI / ML
Object Detection
Training Pipeline
Evaluation
Backend / Serving
Frontend
Model Training Strategy
The classification model was trained using a 4-phase progressive fine-tuning strategy — a technique that allows a pre-trained backbone to be carefully adapted to a new domain without catastrophic forgetting.
Backbone Freeze — Classifier Training
The EfficientNet-B2 backbone weights are fully frozen. Only the custom classification head (Dropout + Linear) is trained, allowing it to learn the waste-domain decision boundary from scratch without disturbing the pre-trained visual features.
Unfreeze Last Block — Fine-Tune
The final feature block of the backbone is unfrozen, allowing the model to adapt its highest-level abstract features to waste-specific patterns while the earlier layers remain stable.
Unfreeze Two Blocks — Deeper Fine-Tune
A second backbone block is released with a further reduced learning rate, refining mid-level texture and edge features critical for distinguishing materials like glass from plastic.
Unfreeze Three Blocks — Full Convergence
The deepest layer of fine-tuning, with a minimal learning rate to make micro-adjustments to low-level visual features. The model stabilises toward its final accuracy on the TrashNet test set.
Future Vision
Where REGAIA Is Going
We believe AI-assisted sustainability should be as ordinary as checking the weather — a quiet, invisible layer of intelligence woven into everyday life.
- Expanded classification categories — moving beyond the six TrashNet classes to include e-waste, batteries, composite packaging, biohazards, and region-specific waste streams.
- Location-aware guidance — integrating local council recycling rules so that, after classification, REGAIA tells users exactly which bin or facility to use based on their postcode or municipality.
- Real-time live detection — a dedicated camera mode that continuously identifies waste items through the viewfinder with bounding boxes and live labels, suitable for smart bins or sorting conveyor use.
- Community impact dashboard — tracking aggregate classifications over time to give users, schools, and organisations a sense of their collective environmental contribution.
- Multilingual support — making the platform accessible to users worldwide regardless of their language, removing another barrier to participation.
- Edge deployment — packaging the model for on-device inference via ONNX or TensorFlow Lite so it can run offline on smart devices, embedded systems, or IoT bins without cloud dependency.
REGAIA begins as a university project, but the problem it addresses is urgent and global. The architecture is deliberately modular — the training pipeline, the inference server, and the frontend are each independently replaceable — so the system can grow without being rebuilt from scratch. The foundation is already sound. What comes next depends on what the planet needs.