For the Great Indian Army - One Recording, Several Readings: Checking AI Detections from Radar, Thermal Cameras and Fibre-Optic Sensors Against the Sensors' Own Settings, for Border Surveillance

Border surveillance uses radars, thermal cameras and fibre-optic sensors, and AI programs now help to call out what these sensors record: a drone, a person, an intrusion. Today the AI gives one detection and one score from one way of processing the recording, and nothing tells the operator how firm that detection is. This concept note describes a simple check that sits on top of any such program, for each of the three kinds of sensor. From the recording the sensor already holds, the computer makes two to four versions by changing only the sensor's own processing settings, within the range the maker permits. These are not invented changes such as flips or artificial noise, so the uncertainty is real. The AI program reads every version unchanged, and one number, the spread value, shows how much the answers disagree. A wide spread means the detection is close to a tipping point, where a small change of setting changes the call. A computer-decided shunting action follows, without waiting for anyone: the detection that needs care is moved to the front of the operator's list, while all other detections carry on. The decision on each detection stays with the operator. No new hardware and no change to the sensor or to the AI program are needed. The note also refers to the Indian Army's AI roadmap and DRDO's multi-sensor border network, and sets the method against current work on radar, thermal and fibre-optic detection. No results on real sensor recordings are given in this note. Patent pending.