Probabilistic neural networks for infrared imaging target discrimination

Abstract

The next generation of infrared imaging trackers and seekers will allow for the implementation of more sophisticated and smarter tracking algorithms, able to keep a positive lock on a targeted aircraft in the presence of countermeasures such as decoy flares. Pattern recognition algorithms will be able to select targets with the help of features extracted from all possible targets images observed in the missile's field of view. Artificial neural networks provide an important class of such algorithms. In particular, probabilistic neural networks are able to reach performances similar to those of optimal Bayesian classifiers by approximating the probability density functions of the features of the samples used in their training. These neural networ ks also present the advantage of generating an output that indicates the confidence level that it has in its answer. We have endeavored to evaluate the performances and the possibility of integrating such neural networks in an infrared imaging seeker emulator, devised by the Defense Research and Development establishment at Valcartier, for countermeasure studies. We describe, in this article, the characteristics that are extracted from the images and define translation invariant features from these. We then give a mathematical basis for the selection of which features to use as input for the neural network, in order to optimize its discriminating power. Finally, we build the neural network and test it on some real data. The results of recognition tests are shown, which indicate a remarkable efficiency: our neural network has a success rate of over 98%. An examination of the images on which it makes its mistakes shows that, for most, even a human expert would probably have also been mistaken. We then build a reduced version of this neural network, with 82% fewer neurons, but with only a 0.6% less precision. We show that such a neural network could well be used in a real time system by measuring its computing time on a normal Pentium 4 PC and obtaining a rate of over 5,300 patterns per second.

Infrared target-flare discrimination using a ZISC hardware neural network

“…The last column gives a reference to that technology However, because the focal plane array imaging seekers represent the latest developments in IR missile seeker technology, no information is publicly available about the particular image processing and pattern recognition algorithms they use. The present study can be seen as the continuation of that in Cayouette's MSc thesis [7] and Cayouette et al [8] that considered the same aircraft-flare discrimination problem we deal with here. This former study used a probabilistic neural network [66] to classify the patterns, seen in a single infrared video frame, as corresponding to aircrafts or flares.…”

Multispectral image analysis for object recognition and classification

“…The features selected for this work were not typical or commonly used features such as SIFT, FAST, SURF, Hu or Hough Moments discussed in Section 2.2, but rather a unique set of features that were previously used in another classification study using thermal imagery only [28]. In Cayouette et al's study, classification rates between 95% and 99.43% were achieved using only two classes of objects (aircraft and flare) that clearly differentiated from each other.…”

“…Cayouette, Labonté and Morin [28] [29] continued the work previously started with Cayouette [28] on the use of PNN to discriminate target aircraft and flares in static images. In the follow-up study, Labonté and Morin considered the time evolution of the image features from a series of frames.…”

<title>Probabilistic neural network, for infrared target discrimination using their temporal behavior</title>

“…More specifically, we explore the ability of an artificial neural network to discriminate between targets. In a first study, Cayouette, Labonté and Morin [4] have shown that a neural network that uses only the static features of the scenes, i.e. those measured in a single video frame, can identify aircrafts and flares with a success rate of over 95%.…”