Date Posted: 05-Feb-2020
Author: Carlo Munoz, Washington, DC
Publication: Jane's International Defence Review
Programme engineers at infrared (IR) sensor and detector manufacturer SCD are in the midst of prototype technology work designed to infuse artificial intelligence (AI) and machine learning (ML) capabilities into advanced versions of their short-wave IR (SWIR) and medium-wave IR (MWIR) sensors, designed to overcome bandwidth congestion issues. The prototype work, conducted under the SCD’s ‘Smart Imagery Consortium’ initiative, is focused specifically on developing an algorithm “that builds the bridge between the focal plane array and artificial intelligence platforms” through advanced variants of the company’s low noise Read-Out Integrated Circuit (ROIC) technology, SCD’s vice-president for Marketing & Business Development, Kobi Zaushnizer, said. “Looking two years, three years, and five years out, I think all of us understand the next generation of imaging must be integrated into the artificial intelligence philosophy. We came to the conclusion that says we need to be ‘AI squared’ [or] artificial intelligence and infrared,” Zaushnizer told Jane’s on 23 January. “In order to enhance the AI, we provide – in the focal plane – an event detection [capability] that enables artificial intelligence applications to do their job faster and easier,” he added. Building upon the company’s work on the ROIC aboard its Cardinal 1280 SWIR platform, engineers looked to tackle two main issues concerning electro-optic and IR (EO/IR) sensor technology: overcoming constrained bandwidth capabilities exacerbated by the proliferation of
high-definition video and imagery, and the ability for EO/IR sensors to detect laser signals – such as laser targeting designators – under adverse weather conditions. “When you go more and more to machine [learning] to do AI, what you need more and more of is a high frame rate ... to be able to detect an event and to be able to see how it will affect your algorithm,” Zaushnizer said. “On the other side there is a [growing] demand for large focal plane array” capability aboard SWIR or MWIR sensors and detectors, to address the laser signal detection issue, he added. “We moved to high definition [imagery], we are moving to 4k ... and the [entire] industry is going in that direction, for more and more resolution, more and more pixels and putting more and more pixels on the target. The issue is when you need to [transfer] a lot of frame rate with very high
resolution, you need a very strong [bandwidth] pipeline,” Zaushnizer said. The AI-enhanced event detection prototype capability being developed by SCD will help overcome
bandwidth bottlenecks, while still maintaining the high-resolution and frame rate for a given image or video. “We can with this [new] architecture ... we can do an automatic event detection inside the focal plane array, with reduced bandwidth,” he said. The new AI-based technology being developed by SCD will “enable a very smart communication
[link] between the application and the sensor ... in kilobytes, instead of megabytes or hundreds of megabytes” with virtually no delay or degradation in the data being transmitted, Zaushnizer said. The other major issue is how various EO/IR or FLIR-based sensors “see” laser targeting designators under adverse conditions. The focal plane array event detection capability being developed under the Smart Imagery initiative can enable SCD sensors to see those laser events without the need for human interaction. The detection capability “knows exactly which pixel” to identify and track the laser event being viewed by the sensor, regardless of any atmospheric or weather-based obstructions. Specific elements from early versions of the AI-enhanced event detector capability created in 2019 have been integrated into SCD-built sensor and detector platforms, with an advanced, full prototype version slated to be complete by the end of 2021. The focus of that prototype work is drilling down into shrinking the event detection capability into an application-specific integrated circuit (ASIC) and integrating it onto the company’s ROIC aboard current and future SWIR sensors, Zaushnizer
said. The Smart Imagery Consortium initiative is part of SCD’s larger effort to expand its product lines and offered EO/IR capabilities over the next 10 years. Dubbed ‘SCD FAB 2020’, the 10-year initiative is intended to support the growth of SCD’s product lines over the next decade by expanding both production and research and defence (R&D) infrastructure for SWIR, MWIR, and LWIR product lines, Zaushnizer said.
