The use of morphology and fuzzy set theory in FLIR target segmentation and classification

The traditional approach to FLIR ATR is detection, segmentation of a silhouette, feature selection, and classification of the target as one and only one type. However, both the infrared data and the target recognition process are not sharply defined, i.e., fuzzy. It is proposed to represent this fuzzy ATR by three techniques: morphology, fuzzy sets, and the low frequencies of the Fourier transform. The target and its internal hot spots are segmented out from the background by use of an iterative morphological contrast peak extraction routine; it uses a new criterion to decide when there is a segmented region, based on differences in the volumes of successive slices of the contrast peak rather than the slope. This routine is more robust than gradient methods for noisy images. The segmented regions can be represented as pseudo fuzzy numbers. Their alpha-cuts represent a set of silhouettes for each segmented blob rather than just one silhouette. For the target or foundation segment the primary recognition feature, silhouette shape, is captured by the low frequencies of the 2-D DFT of each of the alpha-cuts. The hot spots are represented both by the shape features (DFT) and by structural features (e.g., relative position of the exhaust pipe). The first level of this new hierarchical classification system uses an Euclidian distance figure of merit for the foundation's silhouette to assign a fuzzy classification to the target. This initial guess is then adjusted based on the internal features. The fuzzy ATR system is tested on 92 FLIR images containing 3 targets (tank, APC, truck) at 4 aspect angles at about 1.1 km range. The hierarchical system classifies 98.5% of the targets correctly. In addition, 43.6% of the truck's fuzzy measurements are increased in certainty, using the truck's vertical exhaust pipe as an internal feature. Thus this system takes fuzzy FLIR data, creates a fuzzy segmentation, forms fuzzy features, and gives a fuzzy hierarchical classification based on both the shape of the target's silhouette and the shape and structure of the internal hot spots.