Nanomaterial Classifier

Based on the dataset published by Ghada Dahy, Mona M. Soliman, Heba Alshater, Adam Slowik, and Aboul Ella Hassanien

📄 Optimized deep networks for the classification of nanoparticles in scanning electron microscopy imaging, Computational Materials Science, 2023.

---

What This Model Does

This model classifies transmission electron microscopy (TEM) images of palladium nanoparticles (Pd NPs) into six structural categories:

• 0: Non Structure
• 1: Bright
• 2: Circle
• 3: Lone
• 4: Sheet Bends
• 5: Sheet Borders

The goal is to support researchers in identifying nanoparticle morphologies automatically.

---

How It Works

• Input: A TEM image uploaded through Jaqpot.
• Preprocessing: The image is resized to 224x224 and converted to a tensor.
• Model Architecture: Based on ResNet-18 (pretrained on ImageNet), fine-tuned for 6-class classification.
• Output: Raw logits from the model, and the predicted class label.

---

Dataset

• Each class was stored in a separate folder.
• Dataset was split into:
  • 70% training
  • 15% validation
  • 15% testing
• Loaded using torchvision.datasets.ImageFolder.

---

Training Details

• Framework: PyTorch
• Epochs: 10
• Optimizer: Adam
• Loss Function: CrossEntropyLoss
• Selection Criterion: Best validation accuracy
• Evaluation: Accuracy measured on test set

---

Deployment

• Model exported to ONNX format
• A custom ONNX image preprocessor resizes and reshapes the input
• Deployed on Jaqpot via the jaqpotpy SDK

Outputs:

• result: Predicted class label
• class: Raw logits (as string)

---