Auto-Fluidics

Python toolkit for automated droplet detection and flow regime classification in microfluidic experiments using computer vision and machine learning.

Quick Start

git clone git@github.com:FinOrr/auto-fluidics.git
cd auto-fluidics
pip install -r requirements.txt
streamlit run app.py

Features

• Droplet detection using Hough Circle Transform
• Flow regime classification: DRIPPING, JETTING, CO_FLOW, NO_FLOW
• ML classifier using MobileNetV2
• Real-time metrics: count, size distribution, uniformity, aspect ratio
• Interactive Streamlit dashboard and batch processing
• Processing speed: 10-100ms per image

Usage

Dashboard

streamlit run app.py

Python API

from processing.particle_detector import ParticleImageProcessor
from perception.regime_detector import RegimeDetector

processor = ParticleImageProcessor(um_per_pixel=2.0)
processor.process_image('img/sample-10.png')

detector = RegimeDetector()
regime, confidence, _ = detector.detect_regime(processor.particle_metrics)

print(f"Particles: {processor.particle_metrics['num_particles']}")
print(f"Regime: {regime.name} ({confidence:.0%})")

Key Files

Main:

• app.py - Streamlit dashboard
• validate_detection.py - Batch validation

ML:

• train_regime_classifier.py - Train ML classifier
• label_training_data.py - Label training images
• check_ml_setup.py - Verify TensorFlow setup

Video:

• batch_video_extract.py - Extract frames from videos
• extract_training_from_video.py - Extract and label video frames

Core:

• processing/particle_detector.py - Droplet detection engine
• processing/channel_detector.py - Channel ROI extraction
• perception/regime_detector.py - Rule-based regime classification
• perception/ml_regime_classifier.py - ML-based classification

How It Works

Detection: Hough Circle Transform with adaptive parameters, watermark filtering, and outlier removal.

Classification: Rule-based uses particle count, aspect ratio, and size uniformity. ML classifier uses MobileNetV2 fine-tuned on labeled microfluidic images.

License

MIT License - see LICENSE