A-EM (Aerosol Electron Microscopy Analyzer)

A-EM (Aerosol Transmission Electron Microscopy Analyzer) is a Python platform for processing and analyzing aerosol TEM/SEM images. It provides modular, strategy-based pipelines for particle segmentation and quantitative characterization.

Features

• Flexible Architecture — Registry-based plugin system for segmentation backends and analysis engines.
• Multiple Segmentation Backends:
  • Traditional CV (GrabCut, adaptive thresholding)
  • SAM Auto-Mask (Segment Anything Model for improved detection)
• Soot Analysis Engine — Specialized fractal dimension (Df) and morphological characterization.
• Automated Preprocessing — CLAHE, black-hat background removal, denoising.
• Batch Processing — Analyze multiple images in one run.
• Comprehensive Reporting — CSV data export, labeled masks, and visualization plots.
• HyperSpy I/O — Read DM4 and other EM formats with automatic scale calibration.

Installation

pip install a-em

Optional: SAM segmentation support

For GPU (requires CUDA-capable PyTorch):

pip install a-em[sam]

For CPU-only:

pip install a-em[sam]
# Then install CPU-only PyTorch manually:
pip install torch --index-url https://download.pytorch.org/whl/cpu

Quick Start

from a_em import (
    PipelineConfig, PipelineExecutor, HyperSpyReader
)

# Load a DM4 image
signal = HyperSpyReader.read("path/to/image.dm4")

# Configure the pipeline
config = PipelineConfig(
    segmentation_backend="traditional",  # or "sam_automask"
    analysis_engine="soot",
    output_dir="results/",
)

# Run the pipeline
pipeline = PipelineExecutor(config)
aerosols = pipeline.run(signal)

print(f"Detected {len(aerosols)} particles")

Project Structure

a_em/
├── a_em/          # Core package
│   ├── __init__.py
│   ├── config.py           # Pipeline configuration
│   ├── core.py             # AerosolObject model
│   ├── io.py               # HyperSpy-based image I/O
│   ├── preprocess.py       # Image enhancement
│   ├── pipeline.py         # Single-image pipeline executor
│   ├── batch.py            # Batch processor
│   ├── reporter.py         # Results export & visualization
│   ├── segmentation/       # Segmentation backends
│   │   ├── base.py
│   │   ├── registry.py
│   │   ├── traditional.py
│   │   ├── sam_automask.py
│   │   └── legacy.py
│   └── analysis/           # Analysis engines
│       ├── base.py
│       ├── registry.py
│       └── soot.py
├── tests/                  # Test suite
├── examples/               # Example scripts
├── data/                   # Data directory (not included)
├── pyproject.toml
└── README.md

Configuration

PipelineConfig supports the following options:

Parameter	Default	Description
segmentation_backend	"traditional"	"traditional" or "sam_automask"
analysis_engine	"soot"	"soot" (extensible)
output_dir	"data/processed"	Directory for results
clahe_clip	3.0	CLAHE contrast limit
clahe_tile	(16, 16)	CLAHE grid size
background_kernel	25	Morphological kernel size
filter_type	"bilateral"	"bilateral" or "gaussian"

Testing

pytest

License

MIT License — see LICENSE.

Citation

If you use A-EM in your research, please cite this repository.