PaperAgbulut & Józsa (2026), Review of Palaeobotany and Palynology
doi.org/10.1016/j.revpalbo.2026.105731Codegithub.com/borax9/
Free, open-source web app for automated palynofacies analysis of microscope images.
✓ Detects AOM, palynomorphs and phytoclasts
✓ Calibrated areas in µm² (10x, 20x, 40x)
✓ Tyson ternary plot, TAI and stratigraphic logs
✓ Single image or batch upload
✓ Runs in any browser, desktop or mobile
1 · Deep learning: a YOLOv26 instance-segmentation model finds and masks every particle.
2 · Python engine: turns the masks into numbers, a Tyson plot, a TAI score and CSV/XLSX reports.
The image is processed at 640 × 640 px. The model draws a pixel-level mask for every particle and labels it AOM, phytoclast or palynomorph, so background pixels are not counted.
Higher confidence keeps fewer but surer detections; lower keeps more. The curves show the precision–recall balance per class. Best mask F1 (0.92) was at 0.357.
The confusion matrix shows what is mixed up. Most AOM errors are particles blending into the slide background or hidden by opaque phytoclasts.
Pixel areas are converted with Area (µm²) = pixels × (µm/pixel)², calibrated per objective. Relative area % = group area ÷ total area of the three groups × 100.
The three percentages are projected onto the ternary triangle. The nearest of Tyson’s nine facies fields (I–IX) is assigned by Euclidean distance.
The engine averages Hue and Value of the pixels inside each palynomorph mask. Darker colour means higher maturity, from immature (yellow) to overmature (black).
Name files with a depth number (e.g. sample_12.jpg). Paly reads it, sorts the images by depth and plots AOM, palynomorph and phytoclast curves.
⚠️ Important notice
Free Demo Version before Launching a Mobile App and you must increase threshold confidence to get better precision.
We do not claim that palynofacies analysis can be fully automated without error. PalyFacies is designed primarily to accelerate the second stage of the workflow, the quantitative analyses performed by our Palynofacies Engine. It is conceived as a fast, real-time screening tool that can be used at the wellsite, even during mudlogging operations. Rather than serving academic purposes, it has been developed to support petroleum exploration operations.
Ready to pay? You can also pay directly:
Proceed to payment →