The US Army is seeking industry input on a faster target recognition and detection capability that utilizes machine learning algorithms.
In a sources sought notice, the service said the new capability should shorten sensor-to-shooter engagement time to aid an attack.
It must also be able to detect generic classes of targets rather than focus on identifying specific targets with the risk of missing one because of insufficiently trained algorithms.
According to the army, current capability requires a large training image database of specific targets, each captured under a comprehensive set of conditions, such as background terrain, target pose, lighting, and partial occlusion.
Such a requirement reportedly limits the ability to detect targets under new, untrained conditions.
With the recently published request for information, the US Army wants a machine learning-optimized target recognition and detection solution for trained and untrained targets in both trained and untrained conditions.
Needed Characteristics
According to the notice, the new solution must have improved reliability so soldiers can consistently depend on it regardless of their mission type or environmental conditions.
Soldiers must also be able to operate or troubleshoot the capability without degrading their performance.
“The software must be architected in a way to enable rapid upgrade to the latest algorithms, preferably in an automated, cyber-secure fleet management fashion,” the request stated.
Interested firms have until January 12, 2024, to respond. Responses must include detailed answers to 32 questions, including about the solution’s ability to track multiple targets moving at maximum speeds.
