Dashboard

Select a mode to begin detection

Image Analysis

Upload static images for high-precision MPN detection.

Live Stream

Real-time detection from camera feed, optimised for RPi.

History

Browse past prediction results and export as CSV.

MPN Guideline

Reference table and interpretation guide for MPN results.

Image Analysis

Upload a photo of the 9-tube rack to get an MPN result

Drag & drop or click to upload

Live Stream

Real-time tube detection via camera

MPN

Pattern –

MPN / g –

95% CI –

Risk –

Detections

Object	Conf	Value

History

Loading records…

MPN Guideline

Most Probable Number reference for 3-3-3 tube method

What is MPN?

The Most Probable Number (MPN) method estimates the concentration of viable microorganisms in a sample using serial dilutions and presence/absence detection. It is widely used for coliform bacteria testing in food and water safety.

The 3-3-3 Tube Method

Nine tubes are inoculated across three dilution levels — three tubes per dilution. A yellow / bubble result indicates a positive tube (microbial growth detected). A clear / no bubble result is negative.

Group 1 — 0.1 g

1 1 0

Tubes 1–3

Group 2 — 0.01 g

1 0 0

Tubes 4–6

Group 3 — 0.001 g

0 0 0

Tubes 7–9

Pattern

P210

MPN = 15 / g

Interpretation Guide

Safe

< 3 MPN/g

Below detection threshold. Acceptable for consumption.

Low Risk

3 – 20 MPN/g

Low contamination. Monitor and re-test recommended.

Moderate

21 – 110 MPN/g

Elevated contamination. Further investigation required.

High Risk

> 110 MPN/g

High contamination. Sample unsafe. Action required.

Full MPN Reference Table

All 40 patterns for the 3-3-3 tube dilution series.

Pattern	Tubes (x-y-z)	MPN / g	95% CI Low	95% CI High	Risk

Settings

Configure camera, performance, and model parameters

Performance & Source

Camera Source

Inference Mode Aim & Capture is smoother and more accurate on slow CPUs (VPS).

Max FPS — 5 FPS Lower FPS reduces RPi CPU load.

Resolution

Camera Correction

Rotation — 0°

Flip Horizontal (Mirror)

Model

Confidence Threshold — 0.25

Inference Backend … NCNN = fast CPU path. PyTorch = slower fallback (ncnn runtime missing).

Fixed-ROI (Jig Mode)

Classify 9 fixed positions instead of detecting tubes For the fixed camera jig only. Removes missed edge-tubes — the reading always yields 9 tubes. Calibrate once; recalibrate if the camera/rack moves.

Calibration Not calibrated

Snaps a reference frame, auto-detects the 9 tube boxes, then lets you drag to fine-tune before saving.

Network

Device IP Address –