Terminal guidance by image matching goes to war
On 17 January 1991 the destroyer USS Paul F. Foster fired a salvo of Tomahawks in the opening barrage of the war, the missile's first operational use. On the final leg the missile was guided by DSMAC, which compared what its camera saw against stored reference scenes of the ground.
Why it matters
Machine scene recognition for terminal guidance worked in combat twenty-one years before AlexNet and forty before target recognition became a subject of debate. The ancestor of the "drone that recognises and homes" is not a neural network but image correlation against a reference set.
The Navy dates the Block II missile's initial operational capability to 1984. Over water it flies on inertial guidance; over land TERCOM refines the course by matching the readings of the onboard altimeter against elevation maps stored in the missile's computer. Terminal guidance comes from DSMAC, the digital scene matching area correlator, which compares images taken by an onboard camera to scenes stored in the missile computer. The laboratory that developed both says the conventional land-attack Tomahawk they enabled has been a key element of the arsenal since the early 1990s, and that both remain in use when GPS is unavailable. TERCOM's predecessor was Goodyear's ATRAN, a radar map-matching system: in August 1952 Air Materiel Command began mating it to the Martin Matador, with a production contract in June 1954. The record is filed under images deliberately, rather than robotics or foundations: this is a computer vision task, matching a frame against a reference set to say where you are. That it was solved without learning, and inside a combat missile, does not change what it is.