Microscopy Particle Size Analysis

Overview

Automated particle size analysis from microscopy images. The pipeline takes a high-resolution microscopy image (brightfield, darkfield, or DIC mode) and produces a complete granulometric report: binary mask, individual particle measurements, size distribution tables, cumulative curves (D10/D50/D90), and Rosin-Rammler fitting coefficients.

Three-stage pipeline:

Image Preprocessing — Otsu binarization + morphological cleanup + watershed separation
Particle Measurement — Equivalent diameter (Feret + Heywood weighted), shape parameters
Distribution Analysis — DIN 66141 / ISO 9276 classification, D-values, Rosin-Rammler fit

When to Use This Skill

Trigger words: particle size analysis, granulometry, particle counting, microscopy image analysis, 粒度分析, 颗粒计数, 显微图像分析, 粒径分布, D50, D90, Rosin-Rammler, 粉体粒径, 颗粒形态

Applicable scenarios:

Analyzing powder/granule particle size from microscopy images
Counting and measuring particles in brightfield/darkfield/DIC images
Computing particle size distributions (D10, D50, D90, Span)
Fitting Rosin-Rammler distribution to particle data
Measuring particle shape parameters (circularity, aspect ratio, etc.)

Prerequisites

Python Environment

The analysis scripts require the following packages:

numpy>=1.21.0
opencv-python>=4.5.0
scipy>=1.7.0
matplotlib>=3.4.0  # optional, for chart generation

Install into the managed Python venv:

# Create venv if not exists
C:/Users/Administrator/.workbuddy/binaries/python/versions/3.13.12/python.exe -m venv C:/Users/Administrator/.workbuddy/binaries/python/envs/default

# Install dependencies
C:/Users/Administrator/.workbuddy/binaries/python/envs/default/Scripts/pip install numpy opencv-python scipy matplotlib

Scale Calibration Parameters

Before analysis, gather the following microscope configuration:

Eyepiece magnification (M1): typically 10x or 15x
Objective magnification (M2): e.g., 4x, 10x, 20x, 40x, 100x
Camera pixel size: physical sensor pixel size in micrometers (check camera spec sheet)

The scale factor is: S = pixel_size / (M1 * M2) [um/pixel]

Quick Start

Basic Analysis

# Default: brightfield, 10x eyepiece, 40x objective, 5.5um pixels
C:/Users/Administrator/.workbuddy/binaries/python/envs/default/python.exe \
    ~/.workbuddy/skills/microscopy-particle-analysis/scripts/particle_analyzer.py \
    particle_image.tif \
    --output-dir ./output

With Custom Parameters

C:/Users/Administrator/.workbuddy/binaries/python/envs/default/python.exe \
    ~/.workbuddy/skills/microscopy-particle-analysis/scripts/particle_analyzer.py \
    darkfield_sample.png \
    --mode darkfield \
    --eyepiece 10 --objective 20 --pixel-size 5.86 \
    --bins 0.5,1,2,5,10,20,50 \
    --charts --save-mask --save-annotated \
    --output-dir ./results

CLI Parameters

Required

| Parameter | Description | |-----------|-------------| | image | Path to microscopy image file (TIFF, PNG, JPG, BMP) |

Imaging Mode

| Parameter | Default | Description | |-----------|---------|-------------| | --mode | brightfield | Imaging mode: brightfield, darkfield, dic |

Scale Calibration

| Parameter | Default | Description | |-----------|---------|-------------| | --eyepiece | 10.0 | Eyepiece magnification M1 | | --objective | 40.0 | Objective magnification M2 | | --pixel-size | 5.5 | Camera pixel physical size (um) |

Equivalent Diameter Weighting

| Parameter | Default | Description | |-----------|---------|-------------| | --weight-feret | 0.5 | Weight for Feret diameter | | --weight-heywood | 0.5 | Weight for Heywood diameter |

Preprocessing

| Parameter | Default | Description | |-----------|---------|-------------| | --kernel-size | 3 | Morphological kernel size (odd) | | --opening | 2 | Opening iterations (noise removal) | | --closing | 2 | Closing iterations (hole filling) | | --no-watershed | off | Disable watershed separation | | --watershed-threshold | 0.5 | Watershed seed threshold (0-1) | | --min-area | 10 | Min particle area (pixels) | | --max-area | 500000 | Max particle area (pixels) | | --blur-size | 3 | Gaussian blur kernel (0=off) |

Distribution Analysis

| Parameter | Default | Description | |-----------|---------|-------------| | --standard | ISO9276 | Standard: DIN66141 or ISO9276 | | --bin-method | log | Bin spacing: log or arithmetic | | --bins | (default) | Custom bin boundaries, comma-separated (e.g., 0.5,1,2,5,10,20,50) |

Output

| Parameter | Default | Description | |-----------|---------|-------------| | --output-dir | ./output | Output directory | | --charts | off | Generate distribution charts | | --save-mask | off | Save binary mask image | | --save-annotated | off | Save annotated image with contours |

Output Files

The tool generates the following output files in the specified --output-dir:

1. `{basename}_particles.csv`

Individual particle data, one row per particle:

| Column | Unit | Description | |--------|------|-------------| | label | — | Connected component ID | | equivalent_diameter_um | um | Weighted equivalent diameter | | feret_diameter_um | um | Maximum Feret diameter | | min_feret_diameter_um | um | Minimum Feret diameter | | heywood_diameter_um | um | Area-equivalent circle diameter | | area_um2 | um^2 | Particle area | | perimeter_um | um | Contour perimeter | | aspect_ratio | — | Min Feret / Max Feret (0-1) | | circularity | — | 4piA / P^2 (0-1) | | convexity | — | Hull perimeter / contour perimeter (0-1) | | solidity | — | Area / hull area (0-1) | | elongation | — | 1 - aspect_ratio (0-1) | | centroid_x, centroid_y | pixels | Particle centroid position |

2. `{basename}_report.json`

Complete analysis report in JSON format, including:

Image metadata and scale configuration
Preprocessing parameters and particle counts
Full distribution analysis results (bins, D-values, Span, Rosin-Rammler)
Summary statistics (mean, geometric mean, harmonic mean, std dev)

3. `{basename}_distribution_charts.png` (with `--charts`)

Distribution charts:

Top: Number (qn) vs Volume (qv) distribution bar chart
Bottom: Cumulative volume distribution curve with D10/D50/D90 markers

4. `{basename}_mask.png` (with `--save-mask`)

Binary mask image (0=background, 255=particle).

5. `{basename}_annotated.png` (with `--save-annotated`)

Original image with green particle contours and scale bar overlay.

Workflow

When processing a microscopy image for particle analysis:

Gather microscope configuration — determine eyepiece/objective magnification and camera pixel size.
Install dependencies — ensure numpy, opencv-python, scipy are installed in the Python environment.
Run the analyzer — execute particle_analyzer.py with appropriate parameters:
- Set --mode based on imaging technique
- Set --eyepiece, --objective, --pixel-size for accurate physical measurements
- Adjust --bins if custom size ranges are needed
- Enable --charts for visual output
Review results — check the console summary, CSV data, and JSON report. Verify:
- Particle count is reasonable (not 0, not excessively high)
- D50 value is in expected range for the sample
- Rosin-Rammler R² > 0.8 indicates good fit
Adjust parameters if needed — if results are unsatisfactory:
- No particles detected → try --mode darkfield, lower --min-area, or --no-watershed
- Over-segmentation → increase --watershed-threshold (e.g., 0.7)
- Under-segmentation → decrease --watershed-threshold (e.g., 0.3)
- Too many noise particles → increase --min-area or --opening

Algorithm Details

For detailed technical documentation on the algorithms used, refer to:

references/standards_reference.md — DIN 66141 / ISO 9276 standards, equivalent diameter methods, Rosin-Rammler fitting parameters
references/methodology.md — Otsu algorithm, morphological operations, watershed segmentation, Feret diameter calculation, Rosin-Rammler linearization

Script Architecture

scripts/
├── particle_analyzer.py      # Main CLI entry point - orchestrates the full pipeline
├── image_preprocessing.py    # Image loading, Otsu binarization, morphology, watershed
├── particle_measurement.py   # Connected components, Feret/Heywood diameter, shape parameters
├── distribution_analysis.py  # Size classification, D-values, Rosin-Rammler fitting
└── requirements.txt          # Python dependencies

Limitations

Volume estimation: Particle volume is approximated as equivalent sphere (π/6 * D³). For highly non-spherical particles, this introduces bias.
Watershed separation: Cannot perfectly separate heavily overlapping particles. Very tight clusters may still be counted as single particles.
Otsu thresholding: Performs poorly on images with extreme non-uniform illumination. Consider pre-processing with CLAHE in such cases.
Rosin-Rammler fit: Best suited for ground/milled powders. Bimodal or log-normal distributions may show poor R².
Minimum particle size: Particles smaller than ~3 pixels in diameter cannot be reliably measured. Use higher magnification for sub-micron particles.