Threat Fusion Dashboard

Interactive multi-source threat fusion calculator using Dempster-Shafer combination rule. Add drone, HUMINT, SIGINT, and satellite observations to compute fused confidence.

Interactive: Multi-Source Threat Fusion Dashboard

Multi-Source Intelligence Fusion for Drone Warfare

The Dempster-Shafer fusion calculator solves a critical problem in drone-based intelligence: single-sensor observations produce incomplete confidence scores, typically 60-85 percent. A single Fischer 26 detecting a vehicle at 72 percent confidence does not justify committing a 400-euro FPV strike drone. But when three independent drones observe the same location — one at 72 percent, another at 68 percent, and a third at 81 percent — the combined assessment reaches 98.3 percent certainty, meeting the criteria for engagement recommendation. The mathematical formula is simple: C_fused = 1 - ∏(1 - C_i). Each additional observation reduces uncertainty multiplicatively.

Adding Non-Drone Intelligence Sources

The dashboard supports six source types beyond drone observations: HUMINT reports from local informants or captured prisoners, SIGINT intercepts of enemy radio communications, OSINT from social media and open sources, satellite imagery from commercial providers, ground observer reports from forward units, and acoustic sensor detections. Each source type has different reliability characteristics — a HUMINT report might have 50 percent confidence but provides context no camera can capture, such as troop morale or planned withdrawal timelines. Fusing a 50 percent HUMINT report with three 70 percent drone observations produces 97.3 percent confidence, significantly more actionable than drones alone.

Confidence Thresholds and Decision Authority

Lisa 26 uses fused confidence to determine decision authority level. Below 70 percent, the detection is L1 (display on COP only, no action recommended). Between 70 and 85 percent, the system generates an L2 recommendation that a human commander must approve. Above 85 percent with an inbound drone threat and less than 10 seconds to impact, the system may take L3 autonomous action — launching an interceptor without human approval because the human decision cycle exceeds available reaction time. These thresholds are configurable per unit standing operating procedures and rules of engagement.

Read the full technical documentation →

Source Reliability Weighting in Dempster-Shafer

Not all intelligence sources have equal reliability. A Fischer 26 with calibrated IMX477 optics and YOLOv8 trained on Nordic terrain produces higher baseline confidence than a HUMINT report from an unverified informant. The fusion dashboard allows operators to set individual source confidence levels before fusion, reflecting this reality. A drone observation at 72 percent from a proven sensor carries more weight than a 72 percent estimate from an untested source. The mathematical framework handles this naturally — lower-confidence inputs contribute less to the fused result because their uncertainty term (1 - Ci) is larger, reducing their multiplicative impact on the fusion product.

Spatial Clustering and Temporal Decay

The fusion engine only combines observations that refer to the same physical target. Lisa 26 uses spatial clustering with a 100-meter radius — detections within 100 meters of each other are assumed to be the same target and fused together. Detections more than 100 meters apart create separate tracks. Additionally, observations decay over time — a detection from 24 hours ago contributes less confidence than one from 5 minutes ago. The decay function reduces effective confidence by 10 percent per hour after the first 30 minutes, reaching a floor of 20 percent of original confidence after 8 hours. This prevents stale intelligence from inflating the fused picture and ensures the COP reflects current battlefield reality rather than historical positions.

Fusion Output and Automated Threat Assessment

The dashboard classifies fused results into four threat levels. Below 50 percent fused confidence, the contact is labeled UNCERTAIN and displayed as a dashed symbol on the COP — it exists but may be a false positive. Between 50 and 70 percent, the contact is POSSIBLE TARGET with a solid but dim symbol. Between 70 and 85 percent, it becomes PROBABLE TARGET with a highlighted symbol and an L2 engagement recommendation generated automatically. Above 85 percent, the contact is CONFIRMED TARGET with a bright symbol, audio alert to the commander, and a detailed strike package recommendation including recommended FPV type, approach heading, and estimated cost per kill.

Implementation

# Dempster-Shafer Fusion — Multi-Source Confidence
def dempster_shafer(confidences):
    """C_fused = 1 - prod(1 - Ci) for independent observations."""
    miss = 1.0
    for c in confidences:
        miss *= (1.0 - c)
    return round(1.0 - miss, 4)

# Three drone observations of same target
obs = [0.72, 0.68, 0.81]
fused = dempster_shafer(obs)
print(f"Individual: {obs}")
print(f"Fused:      {fused}")  # 0.9830
print(f"Level:      {'L2 RECOMMEND' if fused > 0.70 else 'L1 MONITOR'}")

# Add HUMINT source
obs_with_humint = obs + [0.50]
fused2 = dempster_shafer(obs_with_humint)
print(f"+ HUMINT:   {fused2}")  # 0.9915

Sources

• ArduPilot — ardupilot.org
• FOI — FOI publications catalogue (foi.se/publications)