Spotting a flying object in a country’s skies is very important. While ensuring a safe operation of flights, it will also help spot unauthorised movement of aircraft. Scientists at the International Institute of Information Technology (IIIT-Hyderabad) have developed an IT solution,using artificial intelligence and machine learning tools,for tBharat Electronics Limited (BEL).
It can help Air Force enhance its accuracy in object identification by up to 96 per cent from the present 91 per cent. The software has been transitioned to BEL and is currently being tested out in their simulation environment. The solution, however, doesn’t replace the existing tracking system. “It sits as a layer above the existing system and enhances its ability to spot the aircraft,” Praveen Paruchuri of the Machine Learning Lab at IIIT-H told BusinessLine.
It removes certain redundancies and anomalies in the present system by crunching a large pile of data. The Multi-Sensor Tracking (MST) mechanism, based on a network of radars, creates an Air Situation Picture (ASP), using the data points on location and velocity generated by the radars.
The ASP gives a clear picture of all the aircraft flying in the skies, along with their corresponding flight numbers and flight paths. “The area covered by radars are sometimes overlapped. This leads to a situation where an aircraft is spotted by more than one radar, leading to possible errors,” he said.
“A single aircraft is sensed as multiple aircraft and erroneously flagged as a threat,” he said. To address this challenge, BEL has sought an automated solution from the IIIT-H research team.
The two-member team, comprising Paruchuri and a student Anoop Dasika, trained a machine learning model with 11 days of anonymised and labelled data collected from 17 million data points captured by various radars. The ML model could address the errors, remove redundancies, and improve accuracy by five percentage points.
“Instead of the MST system directly transmitting information to the operators’ screens, now the data from MST goes into our AI system. It will be pre-processed and the original global track number is identified before relaying it back to the original server,” Anoop Dasika said.
Unlike other AI models developed based on a one-time training of data, this model feeds data and trains the model every fortnight.
As a by-product, the ML model can also spot a radar generating more errors, helping the operators go for a repair or replacement.
Findings of the research will be presented at the 34th Annual Conference On Innovative Applications Of Artificial Intelligence (IAAI-22), which will be held virtually from February 24 to 26.