Bespoke object recognition

Your world.
Recognized.

A proprietary recognition engine and custom model layers, specified around your objects, your sensors and your conditions. Detect, track, identify and log where the work happens.

§ 1.1

Hierarchical identification

Platform, type, variant, markings. Defined with you, not mapped onto generic labels.

§ 1.2

Multi-band inputs

Visible, thermal and multi-band imagery with cross-band agreement.

§ 1.3

Calibrated uncertainty

Confidence, an explicit unknown class and decoy assessment on every result.

§ 1.4

Edge, server, hybrid

Recognition held locally when links are denied. Reconciled when they return.

Recognize what matters in your environment. Vektra builds the engine, models and application layers that turn visual inputs into usable intelligence.

From scene to understanding

Built for the details
that define your world.

Custom recognition starts with your objects, your conditions and the decisions your teams need to make.

01 / INFRASTRUCTURE
Substation assets arranged across a utility facility POWER TRANSFORMERPOWER TRANSFORMERCONTROL BUILDING

Utility & infrastructure audits

Identify components, organize asset inventories and connect visual inspection to maintenance workflows.

Explore utility recognition
02 / ASSET INTELLIGENCE
Freight vehicles and containers in a logistics yard TRACTOR-TRAILERSERVICE VANCONTAINER STACK

Vehicles, assets & operations

Build a consistent view of equipment, vehicles and operational activity across complex environments.

Explore asset recognition
03 / DEFENSE
Nadir drone view of two netted 2A65 Msta-B towed howitzers with their crews and a BMP-2 infantry fighting vehicle along a treeline beside plowed farmland 2A65 MSTA-B · 0.91 · NETTED2A65 MSTA-B · 0.88 · NETTEDBMP-2 · IFV · 0.84

Weapons, platforms & units

Identify platforms, weapon systems, variants and unit indicators from drone, aircraft and satellite imagery, under contested conditions.

Explore defense recognition
The Vektra engine

Proprietary at the core.
Specific at every layer.

A recognition system should understand the distinctions that matter to you. We combine a proprietary engine with custom models and application logic built around your operating context.

  • 01Recognition classes defined for your domain
  • 02Accuracy engineering focused on missed detections
  • 03Structured outputs for your existing workflows
Built and running

Detection you can audit.

Independent detectors, a verification layer, visual search with geometric proof and an evidence chain a reviewer can defend, running today in our recognition workspace.

See the capabilities
Multi-model

Independent detectors, compared openly.

Run alone or in tandem, with one-to-one agreement and every model’s raw evidence retained.

Verification

A second look at original resolution.

Candidates are re-examined by a separately trained verification layer, with scores kept apart.

Evidence

A record that holds.

Hashed at upload, immutable once observed, reviewed by a person and written to an append-only audit trail.

Deploy on your terms

Recognition, wherever
you need it.

On the airframe, in the enclave, or both. The same engine runs across every model, so a capability proven on one deployment moves to the next.

Intelligence where the work happens.

Recognition, tracking and identification run on the platform itself: a drone, a vehicle, a handheld or a site gateway. Nothing leaves the edge unless you decide it should.

ComputeNVIDIA Jetson Orin-class modules running TensorRT engines compiled for the device, on the airframe, vehicle or site
ConnectivityNone required. Operates through jamming and outages
DataImagery and results stay on the device under your retention rules
Typical useDrone ISR lock-on, vehicle-mounted recognition, remote substations, airfield aprons

Local inference / Site-level control / Denied-link operation

VISUAL INPUTSEOIR / THERMALFULL-MOTION VIDEOSATELLITE
ON THE PLATFORMVektra engine
DetectTrackIdentifyEvidence
Custom models + application layers
Local identification and track
CENTRAL SERVICESEvidence and model registryEvidence sync when a link is available
Designed behaviour, shown as a diagram.
Where generic recognition fails

Scale, shift, clutter, concealment,
sensor physics and the link.

Most recognition failures are predictable, and we wrote down why: what the public record says about each cause, and what we design against it.

Read the field note
A model built around you

Define it. Train it. Deploy it.

01 / DEFINE

Your recognition problem.

Specify the objects, distinctions and conditions that determine success in your environment.

02 / ENGINEER

Your model layers.

Develop and evaluate domain-specific recognition against representative data and agreed acceptance criteria.

03 / DEPLOY

Your operational fit.

Integrate the engine where it belongs, then refine it through evidence, review and controlled updates.

Built around your challenge

What do you need
to recognize?

Start a project