Multispectral imaging × Edge AI

Machine vision,out in the field

DotData builds vision systems for equipment makers, retail stores and production lines — from choosing the light and the imaging module to training the model and running it offline on a small board. Not a demo: a system people use every day.

400–950 nmVisible to near-infrared
OfflineImages processed on the device
Module to deviceOne team, end to end
Light and opticsNarrow-band LEDs, strobe sync, ring lighting
01
Multispectral captureDark frame, flat field, bad pixels, ambient light
02
Edge AI inferenceSegmentation, detection and tracking on Raspberry Pi-class hardware
03
Decisions and interfaceMachine actions, store app, reports
04
Applications

Vision systems in the field and in development

Every project starts from the same foundation: get the light and the capture right, then put the model on the hardware where the work happens.

Coffee and agricultureResearch

Coffee Bean Sorting

Counts and tracks every bean on a moving conveyor in real time, with multispectral imaging to spot defective beans.

Conveyor count
93% (370 / 399 beans)
Platform
Raspberry Pi 5
Read the case →
Dispensing equipmentCo-developed with a maker

Medication Recognition

Checks whether a tray cell holds a pill, and finds the best point for the suction nozzle to pick each one.

Internal scenario test
36 / 36 pick points correct
Platform
Raspberry Pi CM4
Read the case →
Salons and skincare storesIn store trials

Skin Analysis

Multispectral analysis of spots, brown areas and red areas. Face photos stay on the in-store host.

Report
≈ 10 s
Interface
iPhone app
Read the case →
How we work

From one question to a machine used every day

You bring the problem and the samples. We tell you whether it can be done and what it will cost, then take it all the way to production.

  1. Feasibility

    We study your samples and measure spectra to decide between RGB, multispectral or another sensing method.

    Deliverable: feasibility report
  2. Prototype

    We build a working prototype and measure accuracy and speed on your real samples.

    Deliverable: prototype + test data
  3. Field deployment

    We install the system in your machine or store and tune it under real lighting and pace.

    Deliverable: acceptance record
  4. Production

    We cut hardware cost, package it as a repeatable module and keep the model updated.

    Deliverable: module / SDK
Vision notes

Lessons from taking machine vision into the field

Short articles from our projects, for equipment makers, brands and engineers evaluating a vision solution.

  • Multispectral
    Why can't an RGB camera tell spots from redness?Starting from melanin and hemoglobin absorption
    Coming soon
  • Capture
    Four steps of multispectral calibrationDark frame, flat field, bad-pixel mask, ambient light
    Coming soon
  • Edge AI
    Running segmentation on a 2 GB boardTrading off quantization, input size and memory
    Coming soon
  • Production
    Counting beans on a conveyorWhy tracking is harder than detection
    Coming soon

Is there a job people still do by eye?

Equipment development, inspection, store deployment, research partnerships or investment — write to us. Send sample photos or a description of the site and we will reply with an initial assessment.

Email
Vision assessmentModule purchaseCo-developmentStore trialInvestment