Electro-Optical AI Detection & Classification

Data status: Illustrative

A camera can see what a radio trace cannot, and it can be fooled by the same things that fool a person: haze, glare, dusk, distance. Seeing that something is there is easier than saying what it is, and a system that blurs the two will sound surer than it is.

• Image framesPassive optical or thermal views of the sky, timestamped. No sensor model, lens or resolution is published here.
• Visibility conditionsLight, haze and contrast around the sensor, which decide how much an image can support.
• Sensor healthWhether the sensor is delivering usable frames, so that a blank view can be told from an empty sky.

1. Looks for something that is distinct from the sky behind it.
2. Separately asks whether the detected thing can be classed, and scores that more strictly.
3. Emits observation objects whose detection and classification confidence are kept apart and tied to visibility.

• Detection and classification are two claims, scored separately.
• Classification is allowed to fail first, and the output says so.
• Visibility is a visible input, not a hidden assumption.
• A class label never travels without its uncertainty.

Electro-optical detection is a source at the start of the chain. It turns images into observations in which detecting and classifying are scored apart, and hands them to correlation and fusion in the core platform. It does not decide what an object is on its own.

Read the architecture

Evidence ladder

1. E0 · Concept reached
2. E1 not reached
3. E2 not reached
4. E3 not reached
5. E4 not reached
6. E5 not reached

Project assurance panel

Mission and context

Add an optical trace to the airspace state and keep clear the difference between seeing something and knowing what it is.

Primary contribution

Observation objects in which detection and classification are separate claims, each tied to visibility.

Decision value

Shows a reviewer when an optical view can say that something is there, and when it can also say what it is.

Evidence status

Evidence level: E0 · Concept Concept and research direction only. No specification, prototype, test result or deployment is claimed.

Next validation step

Write down how detection and classification are scored apart, then specify a first evaluation on synthetic frames before any claim is made.

Behaviour when degraded

As visibility falls, classification confidence drops before detection confidence does, and the observation says so. When the sensor stops delivering frames, the state reports absence of a view, not an empty sky.

Role in integration

Would supply observations to correlation and fusion and sensor health to system monitoring.

Human review

A person reviews any state in which an optical class label contributes to an action.

Public-safe limitation

No sensor, lens, resolution, model or site is described. The slider on this page uses invented curves.

Open unknowns

Which conditions matter most, how classification should be scored and how the two claims should be combined are not yet known.

As visibility falls, classification confidence drops before detection confidence does, and the observation says so. When the sensor stops delivering frames, the state reports absence of a view, not an empty sky.

• A site where a visual confirmation is wanted

  Why it matters here

  A second, independent kind of trace helps to tell cases apart.

  What it contributes

  An optical observation with its limits stated.

  What it does not replace

  A person looking, or a radio or spatial trace.

• Conditions that change through the day

  Why it matters here

  Light and haze change what an image can support.

  What it contributes

  A classification claim that weakens openly as visibility falls.

  What it does not replace

  A decision about what the object is.

Would supply observations to correlation and fusion and sensor health to system monitoring.

Write down how detection and classification are scored apart, then specify a first evaluation on synthetic frames before any claim is made.

• This is a concept. There is no specification, prototype or measured result behind it.
• The slider uses invented curves. They show the relation between the two claims, not the behaviour of a system.
• Classification is shown failing before detection. How much earlier is not known and is not stated.
• An optical view depends on a line of sight and on light, so it says nothing about what is out of view.

• Passive RF Sensing NodeEvidence level: E1

  The passive RF node adds a trace that does not depend on light.

• Passive Coherent Location (PCL) RadarEvidence level: E1

  Passive coherent location adds spatial observations that do not depend on a clear view.

• Multi-Sensor Fusion EngineEvidence level: E0

  Fusion combines the optical trace with the others and keeps their uncertainty.

• Drone Detection & Airspace Intrusion IntelligenceEvidence level: E0

  Low-altitude detection uses optical views as one thread of its evidence.

• Airport Airspace IntelligenceEvidence level: E0

  Airport airspace is a setting where knowing what an object is, and how sure we are, matters.