Discrimination of Aircrafts and Flares in Infrared Images by a Probabilistic Neural Network

P. Cayouette, G. Labonté, and A. Morin (Canada)


Probabilistic neural network, target classification, tracking.


We describe a probabilistic neural network that is very successful at discriminating between infrared images of aircrafts and their decoy flares. Such neural networks present the unique advantage of giving as output the probability that the features observed correspond to an object of any one of many classes. An infrared seeker devised at the Defense Research and Development Canada provides a certain set of image characteristics. From these, we define translation and rotation invariant features. We describe the method we used for the selection of which of these features to include as input for our neural network, in order to optimize its discriminating power. We show the structure of the neural network and exhibit the results of our tests. Our neural network is seen to have a success rate of over 98%. An examination of the images on which it makes its mistakes shows that most of these are such that even a human expert would have been misled. Finally, we report preliminary results about the computing efficiency of a pruned version of our network that can identify over 6,300 patterns per second on an ordinary PC.

Important Links:

Go Back