🔬 AquaTrace AI - Polymer Classification Model
Model Description
This model classifies the type of plastic polymer based on spectral signatures from multispectral satellite imagery. It's part of the AquaTrace AI accountability system.
Key Features
- Architecture: EfficientNet-B0
- Performance: 86.56% accuracy
- Classes: HDPE, LDPE, PET, PP, PS, PVC, PA (Nylon)
- Method: Refractive index analysis + spectral classification
Model Architecture
EfficientNet-B0 modified for spectral input:
- Backbone: EfficientNet-B0
- Input Channels: 9 (multispectral bands)
- Output Classes: 7 polymer types
- Parameters: ~5M
Polymer Classes
| Class | Full Name | Refractive Index | Common Uses |
|---|---|---|---|
| HDPE | High-Density Polyethylene | 1.54 | Bottles, containers |
| LDPE | Low-Density Polyethylene | 1.52 | Bags, films |
| PET | Polyethylene Terephthalate | 1.58 | Beverage bottles |
| PP | Polypropylene | 1.49 | Food containers |
| PS | Polystyrene | 1.59 | Foam, packaging |
| PVC | Polyvinyl Chloride | 1.54 | Pipes, packaging |
| PA | Polyamide (Nylon) | 1.53 | Fishing gear |
Training Data
- Dataset: Synthetic + real spectral signatures
- Training Samples: 10,000+ labeled samples
- Validation Accuracy: 86.56% accuracy
Usage
import torch
import torchvision.models as models
# Load model
model = models.efficientnet_b0(pretrained=False)
model.features[0][0] = torch.nn.Conv2d(9, 32, kernel_size=3, stride=2, padding=1, bias=False)
model.classifier = torch.nn.Sequential(
torch.nn.Dropout(0.3),
torch.nn.Linear(1280, 7)
)
checkpoint = torch.load("best_model.pth")
model.load_state_dict(checkpoint['model_state_dict'])
model.eval()
# Inference
spectral_data = torch.randn(1, 9, 224, 224)
with torch.no_grad():
outputs = model(spectral_data)
predicted = torch.argmax(outputs, dim=1)
classes = ['HDPE', 'LDPE', 'PET', 'PP', 'PS', 'PVC', 'PA']
polymer_type = classes[predicted.item()]
print(f"Detected polymer: {polymer_type}")
Performance Metrics
| Metric | Value |
|---|---|
| Accuracy | 86.56% |
| Precision (avg) | 84.2% |
| Recall (avg) | 83.8% |
| F1-Score (avg) | 84.0% |
Citation
@software{aquatrace_polymer_2026,
author = {AquaTrace AI Team},
title = {AquaTrace AI Polymer Classification Model},
year = {2026},
publisher = {Hugging Face},
url = {https://huggingface.co/Klassy01/aquatrace-ai-polymer-classification}
}
License
MIT License