Architecture

From an RGB photograph to a decision, in six stages.

Nabhya is crop intelligence infrastructure. This is what happens between your upload and the response.

Pipeline

RGB frame (+ optional AOI)
      |
      v
  segmentation  -->  excluded pixels reported
      |
      v
   inference  (CPU, ~2s)
      |
      v
  near-NDVI field
      |
      v
  percentile ranking
      |
      v
 PNG   JSON   GeoTIFF

Stages

01: RGB input; A standard drone photograph, orthomosaic tile or GeoTIFF. Optionally an AOI boundary as GeoJSON.
02: Segmentation; Vegetation is separated from rooftops, roads, water and bright bare soil. Non-crop pixels are excluded from every downstream statistic.
03: Inference; The trained model maps visible-light pixels to a near-NDVI value, frame by frame, on CPU.
04: Near-NDVI field; A continuous per-pixel vigour field aligned to the submitted frame, rendered green through yellow to red.
05: Ranking; Percentiles within the analysed area produce management zones: worst 10%, middle, best 10%.
06: Outputs; PNG panels, a JSON summary with pixel accounting and confidence, and a georeferenced GeoTIFF raster.

Deployment

Hosted API

The default. One HTTPS endpoint, keys per environment.

Batch

Submit a full flight, collect results as they complete.

On-premise / edge

CPU-only inference runs on your own hardware or a ground station. Commercial arrangement.

Get a result

Send us one frame.

If you have a recent RGB flight sitting on a drive, send a single frame and we'll return the full result for it — free, no account. It's the fastest way to find out whether this is useful to you.

Returned: Vigour map, zones, canopy mask, summary
Turnaround: Same day, seconds of compute
Cost: Free during onboarding