Machine Brand Recognition

Case file Early access

Your machine count has no make on it. We're building the read that adds one.

Machine Brand Recognition reads livery colour, boom geometry and counterweight silhouette from very-high-resolution imagery and attributes each detected machine to a manufacturer, with a confidence tier. It is in early access, so it starts as a pilot list.

Join the pilot listSee how it works
GSD
VHR ≤0.3 m
Cadence
Quarterly
Platform
VHR satellite / aerial
Spectral
RGB
Outline of a tracked excavator with three cues marked: livery colour on the house, boom geometry, and the counterweight at the rear A livery colourB boom geometryC counterweight
Schematic, not a real read. The three cues the attribution works from: livery colour (A), boom geometry (B) and the counterweight silhouette (C).

The job today

A count is not a fleet share

You have the number: a count of machines on a site, from a detection run or a survey. The deck needs something else. The quarterly competitive review wants to know how many of those are yours, how many belong to the rival down the road and how many are a third make nobody expected. The count says nothing about any of that.

Black and white photo of tire and track marks crossing a graded site, with a pipe stack and sheeted material at the back
Exhibit 1Track marks across a cleared site.

So the brand mix gets built from scraps. A dealer mentions a rival's fleet at a rental yard. A sales rep photographs a jobsite from the road. Somebody pulls a registration list, where the market publishes one, and a rental company's fleet sheet covers a few of the yards. Each scrap is real and each covers a slice, most of it late. By the time the slide is assembled it carries more caveats than numbers, and a regional manager asks which sites the percentages came from.

Machine Brand Recognition adds the missing column. For each machine already detected on a site, it reads what the imagery shows of the paint and the shape and proposes a make, with a confidence tier so you can see which reads to trust.

What you get

What the pilot would deliver

Three files per site, each one built to drop into the tools an analyst already opens. Because this is early access, the exact form gets agreed with each pilot participant.

  1. 2.1

    Per-machine attribution layer

    GeoPackage or CSV

    One row per detected machine: position, machine class, the proposed manufacturer and a confidence tier. It opens in your GIS or loads into the spreadsheet behind your share charts.

  2. 2.2

    Site brand-mix summary

    CSV

    The same attributions rolled up per site: how many machines per make, and how many unattributed. It is the table that feeds the fleet-share slide.

  3. 2.3

    Review list

    CSV

    Every attribution that came back at a low confidence tier, pulled out so a person checks those before anything reaches a deck. Machines whose make couldn't be read are listed here as unattributed.

How it works

  1. Step 1

    Send the sites

    Give us the sites or fleet areas you want read, as boundaries or coordinates, plus the machine detections you already have if you have them.

  2. Step 2

    Machines detected

    Machines on each site are found in very-high-resolution imagery. If you have no detection layer, we run that step first.

  3. Step 3

    Livery and shape read

    Colour, boom geometry and counterweight silhouette are read for each machine and matched to a manufacturer.

  4. Step 4

    Summary back for review

    You get the attribution layer, the brand-mix summary and the review list. The review list is for your team to check before the numbers go into the competitive review.

Where it fits in your workflow

It sits after the machine count and before the deck. Your base count tells you how many machines are on a site. This read turns that count into share by make, which is the number competitive intelligence reports. It adds to the dealer feedback, the registration data and the rep photos. You would compare it with those sources, not throw them out.

Limits

  1. L1

    Manufacturer, not model

    The read stops at the maker. It doesn't say which model a machine is, and it doesn't read serial numbers or decals.

  2. L2

    Pixels set the floor

    At 0.3 m a pixel is about a foot. A counterweight shows up as a handful of pixels across and a boom as only a few, so livery colour and overall shape are what the read works from. Small machines and machines parked close together, or partly hidden, give it less to go on, and those come back at a lower tier.

  3. L3

    Repainted and unbranded machines

    A machine in a rental company's own colours, a repaint or a machine under a tarp carries no readable make, so it gets flagged as unattributed instead of guessed at.

  4. L4

    Accuracy varies, and we don't publish a figure

    How well attribution works depends on how visible the livery is and what class of machine it is. A blanket number would be made up, so there isn't one on this page. Testing it against a fleet you know is something to raise in the pilot.

  5. L5

    Quarterly, and cloud matters

    The target cadence is quarterly, which lines up with a quarterly competitive review. Cloud and haze over a site can push a pass back.

  6. L6

    Early access

    Nothing here ships as a finished report yet. The pilot list is how we decide what to build first.

Who it's for

For you if

  • You work in competitive intelligence or market analysis at a machinery maker and report fleet share or installed-base mix.
  • You already have machine counts, or want them, and the missing piece is the make.

Not for you if

  • You need the exact model of each machine.
  • You need a guaranteed brand mix or an accuracy figure on day one.
  • You want a finished product today. This is a pilot list.

FAQ

Questions before you join

How accurate is attribution?

We don't publish a blanket figure. Accuracy varies with how visible the livery is and what class of machine it is. Ask about your fleet and raise how to test it on sites you know when you join the pilot list.

What happens to repainted or unbranded machines?

They are listed as unattributed, not guessed at.

Does it identify the exact model?

No. The read is manufacturer-level only.

Why quarterly and not monthly?

A quarterly read lines up with the competitive review. A faster cadence is something to raise in the pilot.

Can I check it against what I already know?

Yes, and you should. Compare the attributions to your dealer data, rental fleet lists or sites you have walked. The review list shows where to look first.

Do I need machine counts first?

It helps. If you have a detection layer, send it. If not, the detection step runs first.

How does it plug into our tools?

The files are standard formats that open in a GIS or a spreadsheet. There is no direct integration with BI or CRM systems.

Who owns the output, and what about the imagery rights?

Your site list stays yours. Rights to the outputs and to the imagery behind them get written into the pilot agreement.

What does it cost?

There is no price on this page. Pricing is scoped per pilot once we know your sites.

Get on the pilot list.

Tell us which sites or fleets you want read and what decision the fleet-share number feeds. Join the list and we'll follow up about the pilot.

Join the pilot list