Accessibility settings

Settings are saved in this browser and apply on every page of the site.

Text size 100%

Accessibility is built into the site code. This menu is an extra, not a substitute.
HAMMER GIS, home
עב Talk to us

HG-S05 SEE.GMMA · by HAMMER

From orthophoto to GIS layer, with AI assistance and human review

SEE.GMMA helps a GIS professional produce city-scale vector layers from large aerial images. The NOW module splits the image into tiles, sends only the tiles you select for AI analysis, and turns the result into vectors with a local engine. The WHEN module searches free satellite archives to find when a place changed.

Pilot

Book a demo NOW and WHEN
Dark aerial photo of hills with AI-mapped orchard and forest polygons
P-01 Dark aerial photo of hills with AI-mapped orchard and forest polygons
Plan no.
HG-S05
Scale
1:1,250
Sheet
05/05
Updated
11.2026
Status
Pilot

The problem

A whole city, one polygon at a time

Digitizing a city from an orthophoto by hand is slow, grinding work: roads, pavements, buildings, parks, kerbs and parking, one after another. And anyone who needs to know when a building went up or a green area disappeared ends up flipping through old photos with no method.

Earlier aerial photo of a low-rise neighbourhood
Later photo: new building footprints marked in red
Earlier Later
P-02 Later photo: new building footprints marked in red

NOW · current state

From aerial photo to vector layers

NOW targets urban surface layers such as roads, pavements, buildings, parks, kerbs and parking.

  • Large orthophotos, in place ECW and SID files open where they are, without copying, showing CRS, extent and resolution.
  • A persistent processing grid Tile selection with sea and blank-tile exclusion and an area of interest.
  • Only the tiles you choose Only explicitly selected tiles are sent to a Google Gemini image model, interactively or in batch, with a cost estimate up front and no paying twice for the same tile.
  • Deterministic vectorization A local engine turns the result into areas and lines, stitches seams between tiles and merges blocks.
  • Export GeoPackage or GeoJSON, in the source CRS or in ITM.
  • An audit trail per tile Every AI result keeps the model, prompt version, tile ID and raw response.
Neighbourhood with green outlines of mapped buildings
P-03 Neighbourhood with green outlines of mapped buildings

WHEN · when did it change

Forty years of satellite archive, answered in two dates

  • Pick a cell Draw it, load SHP, GeoJSON, GPKG or KML, or pick from the tiles.
  • Search back Up to 40 years back: Landsat from 1984 at 30 m, and Sentinel-2 from 2017 at 10 m.
  • Four kinds of change Water, vegetation, built-up or bare soil.
  • Two dates The last date the change had "not yet" happened, and the first date it had "already" happened.
  • Cost and time first An estimate before anything runs.
  • Clear limits WHEN detects changes at lot scale or larger, and a person must confirm every change.
Building footprints in blue with pavement outlines over the orthophoto
P-04 Building footprints in blue with pavement outlines over the orthophoto

How it works

Seven stations from photo to layer

  1. ST-01

    Open the orthophoto

    The file stays where it is.

  2. ST-02

    Grid and selection

    Choose tiles; exclude sea and blank areas.

  3. ST-03

    Cost estimate

    See what the analysis will cost before it runs.

  4. ST-04

    AI analysis

    Only the selected tiles go to the model.

  5. ST-05

    Vectorization

    A local, deterministic engine turns the result into areas and lines.

  6. ST-06

    Review by a GIS professional

    The output is checked and edited before it becomes a layer.

  7. ST-07

    Export

    GeoPackage or GeoJSON, in the source CRS or ITM.

S-05

Principle

AI assists. A person decides.

SEE.GMMA doesn't promise automatic mapping. NOW output is checked and edited by a GIS professional before it becomes a layer, and changes found by WHEN wait for a person to confirm them. The dedicated review tools, including snapit, a browser editor for polishing layers collaboratively, are still in development.

What happens to your imagery

Straight answers about data

  • The full image stays put The orthophoto file opens in place and is not copied.
  • Selected tiles go out for analysis The tiles you choose, and only those, are sent to Google Gemini. Worth knowing before you select.
  • Every result is logged Model, prompt version, tile ID and raw response for every AI result.
Later photo: new building footprints marked in red
P-05 Later photo: new building footprints marked in red

Who it's for

For anyone who needs layers from aerial imagery

  • GIS professionals City-scale vector layers from large aerial images, with control over every tile.
  • Municipalities Urban surface layers: roads, pavements, buildings, parks, kerbs and parking.
  • Anyone tracking land change A documented answer to when an area became built-up, vegetated, water or bare soil.

Integrations and formats

What SEE.GMMA works with

  • ECW / MrSID Orthophoto input
  • Google Gemini Image models, including the Batch API
  • Landsat Archive from 1984, 30 m
  • Sentinel-2 Archive from 2017, 10 m
  • GeoPackage / GeoJSON Export
  • SHP / KML Cell input for WHEN
  • ITM Export in the Israeli grid

Need layers right now?

OrthoGraphic AI service

Separately from SEE.GMMA, HAMMER delivers aerial-image digitizing as a service: machine recognition, manual completion and quality control by a team of image interpreters.

Status

Pilot

SEE.GMMA is in active development and in a pilot with an Israeli municipality. A first full-city run is awaiting approval, and the human-review tools are in development. There is no public sign-up yet, and we don't publish accuracy figures until they have been measured and validated.

S-11

Got an orthophoto waiting to become layers?

In a demo we'll show NOW and WHEN on sample data, and see whether the pilot is a fit for you.

Book a demo