NEURAL NETWORKS FOR AUTOMATIC TARGET RECOGNITION

Many applications reported in artificial neural networks are associate d with military problems. This paper reviews concepts associated with the processing of military data to find and recognize targets-automati c target recognition (ATR). A general-purpose automatic target recogni tion system does not exist. The work presented here is demonstrated on military data, but it can only be considered proof of principle until systems are fielded and proven ''under-fire''. ATR data can be in the form of non-imaging one-dimensional sensor returns, such as ultra-hig h range-resolution radar returns for air-to-air automatic target recog nition and vibration signatures from a laser radar for recognition of ground targets. The ATR data can be two-dimensional images. The most c ommon ATR images are infrared but current systems must also deal with synthetic aperture radar images. Finally, the data can be three-dimens ional, such as sequences of multiple exposures taken over time from a nonstationary world. Targets move, as do sensors, and that movement ca n be exploited by the ATR. Hyperspectral data, which are views of the same piece of the world looking at different spectral band, is another example of multiple image data; the third dimension is now wavelength and nor time. ATR system design usually consists of four stages. The first stage is to select the sensor or sensors to produce the target m easurements. The next stage is the preprocessing of the data and the l ocation of regions of interest within the data (segmentation). The hum an retina is a ruthless preprocessor. Physiology motivated preprocessi ng and segmentation is demonstrated along with supervised and unsuperv ised artificial neural segmentation techniques. The third design step is feature extraction and selection. the extraction of a set of number s which characterize regions of the data. The last step is the process ing of the features for decision making (classification). The area of classification is where most ATR related neural network research has b een accomplished. The relation of neural classifiers to Bayesian techn iques is emphasized along with the more recent use of feature sequence s to enhance classification. The principal theme of this paper is that artificial neural networks have proven to be an interesting and usefu l alternate processing strategy. Artificial neural techniques, however , are not magical solutions with mystical abilities that work without good engineering. Good understanding of the capabilities and limitatio ns of neural techniques is required to apply them productively to ATR problems.