Given advancements in machine vision, what issues would there be if air to air missile guidance used image recognition (at least in the terminal phase) to defeat flares and chaff?

  • 8
    $\begingroup$ Night. Clouds/fog. $\endgroup$ Commented May 18 at 20:30

2 Answers 2


It is used by modern missiles. Military isn't the birthplace for every tech, but they've been working on image recognition long before the general public heard the term "image recognition". This 1989 article, barely 35 years old, elaborates on some of the earliest neural networks used.

Shape is but one of many factors used by missile seekers to differentiate the true target. Dynamics and behavior are just as important. This is because flares are very bright, and the image in question very low-resolution.

Different guidance techniques for the approach phase and the terminal phase is outright 50 years old, since the second generation of contrast seekers, though early systems just went for the image center.

enter image description here

  • 8
    $\begingroup$ Let it not be unsaid that neural networks and computer vision aren’t inseparable; when I published a paper on the latter in 1987 there was already a wealth of techniques and at that time most didn’t use neural networks. $\endgroup$
    – Frog
    Commented May 19 at 20:19
  • 4
    $\begingroup$ Military research into projectiles guided by image recognition appears to date back to the 1940s. $\endgroup$
    – Mark
    Commented May 20 at 23:29

Modern missiles are using computer vision or image processing technology.

Obviously, for this to work you need an image, which is typically provided by an image-generating IR-sensor (note that this does not work with radar sensors of course). In the past, IR-sensors would only try to determine the brightest (hottest) spot in the sky. This technique is of course susceptible to counter-measures such as an IR-flare or trying to get the missile to lock onto the sun. But an image of course contains much more information which can be used to improve guidance and differentiate counter-measures. Current seekers try to use the information contained in the monochrome picture to extract more information and discern countermeasures from the target etc.. In the future multi-band imaging sensors will be used in order to gain even more information, e.g. a color picture vs a monochrome picture.

One good example which uses image recognition today to improve its end-game capability, is the (now famous) Diehl IRIS-T:

Intelligent image processing detects IR decoys in the image and ignores them.[...] For identification purposes, there is a target library that contains images of all known military aircraft from different perspectives. For each target, eight of the most vulnerable points are saved, towards which the guided missile is heading.

Translated to english, from the german IRIS-T wikipedia article (I should note that not all links in that quote are available anymore).

This is known as "aim point selection" which can only be used, if you have a somewhat advanced understanding of the image which your seeker currently sees. Other missiles such as the AIM-9X Sidewinder or the ASRAAM also use imaging IR-Sensors, and most likely employ similar techniques as the IRIS-T.

And this apparently works very well:

Buschek said Ukraine had shot down more than 110 targets, most of these cruise missiles such as the Kalibr, with a hit rate of almost 100%.


You can get a good sense of how such an engagement looks from the missile side in this paper:

enter image description here

  • $\begingroup$ Re, "this does not work with radar sensors" Specifically, it does not work for acquiring and tracking targets for anti-aircraft missiles, but you can use much larger radar antennas to make interesting images of much larger targets. public.nrao.edu/news/radar-tycho-crater-intricate-detail $\endgroup$ Commented May 19 at 23:14

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .