Neural networks for automatic target recognition

Abslract--Many applications reported in artificial neural networks are associated with military problems. This paper reviews concepts associated with the processing o f military data to f ind and recognize targets----automatic target recognition ( A TR ). A general-purpose automatic target recognition system does not exist. The work presented here is demonstrated on military data, but it can only be considered proof o f principle until systems are fielded and proven "'under-fire". A TR data can be in the form o f non-imaging one-dimensional sensor returns, such as ultra-high rangeresolution radar returns for air-to-air automatic target recognition and vibration signatures from a laser radar for recognition o f ground targets. The A TR data can be two-dimensional images. The most common A TR images are infrared, but current systems must also deal with synthetic aperture radar images. Finally, the data can be threedimensional, such as sequences o f multiple exposures taken over time from a nonstationary world. Targets move, as do sensors, and that movement can be exploited by the A TR. Hyperspectral data, which are views o f the same piece o f the worm looking at different spectral bands, is another example o f multiple image data; the third dimension is now wavelength and not time. A TR system design usually consists o f f our stages. The first stage is to select the sensor or sensors to produce the target measurements. The next stage is the preprocessing o f the data and the location o f regions o f interest within the data (segmentation). The human retina is a ruthless preprocessor. Physiology motivated preprocessing and segmentation is demonstrated along with supervised and unsupervised artificial neural segmentation techniques. The third design step is feature extraction and selection: the extraction o f a set o f numbers which characterize regions o f the data. The last step is the processing o f the features for decision making (classification). The area o f classification is where most A TR related neural network research has been accomplished. The relation o f neural classifiers to Bayesian techniques is emphasized along with the more recent use o f feature sequences to enhance classification. The principal theme o f this paper is that artificial neural networks have proven to be an interesting and useful alternate processing strategy. Artificial neural techniques, however, are not magical solutions with mystical abilities that work without good engineering. Good understanding o f the capabilities and limitations o f neural techniques is required to apply them productively to A TR problems.