Principles of Modern Radar - Volume 2 1891121537

656 CHAPTER 14 Automatic Target Recognitionordinate, u [75]. Thus, the log-spiral transform is translation, rotation, and scale invariantwhen the centroid is moved to a baseline point for each image [75].14.5.3.5 Ship FeaturesFeatures for land targets have been previously discussed in the SAR section. However,images of ships and other watercraft at sea are often linked specifically to ISAR imagery;therefore, useful features for ship classification are discussed.Ships at sea are a common target for ISAR imaging with multiple aspect angles andimage orientations captured due to the changing direction of the angular velocity vectoras the ship moves along the waves. Some of the most useful classification features forlarge ships are ship length, number of masts, and mast location [67]. For small ships, thesefeatures can be supplemented with width or height, shape, guns, and gun location [67].Scaled and oriented wire grids from candidate ship libraries can be compared to ISARship images to aid in classification [67, 69]. To determine the orientation of a ship in anISAR image, features such as the centerline (using maximum peak of Hough transform ofpartially segmented image), locations of the bow and stern, how much plan informationis present (peaks in the Hough transform at the angle parallel to centerline), how muchprofile information is present, the width of the plan at either end, and the width of theoutline at either end can be used [69]. In a process sometimes referred to as mensuration,the locations of major structures, such as masts, superstructure breaks, cranes, andweapons, are found along the bow-stern axis and normalized relative to the range extentof the ship; these locations can be determined by looking at the number of target pixelsextending in the Doppler direction beyond either side of the deck edges for each rangecell [69]. Other useful features in ISAR ship imagery include: straight, curved, or roundedstern; pole, lattice, or solid mast; target range extent; percentage of the target range extentcovered by superstructure; number of major uprights; the ratio of the highest mast to themedian height of the profile; and the correlation of the five best profiles with a storedcharacteristic profile shape [69]. Because many of these features require maximum heightprofile information, ship ISAR images are selected for evaluation based on a histogramof the number of target pixels per range cell on the side of the target with the most pixelsoutside of the plan boundary. Vertical streaks caused by transient or rapid motions internalto the target (e.g., moving machinery) must be removed before extracting relevant featuresfrom the image. This can be done by counting the number of pixels above a threshold incross-range for each range cell, by flagging range cells with more than 66% of their pixelsabove this threshold, by finding pixels in these flagged ranges that are bright while theirneighbors are not, and by reducing the intensity of such pixels by 30% [69].14.6 PASSIVE RADAR ATRWhile traditional monostatic radar systems like those commonly used to perform SAR andISAR include both transmitters and receivers, passive radar systems contain only receivers.They rely on “illuminators of opportunity” already present in the environment (e.g., antennasused to transmit signals for television, cell phones, or FM radio) to illuminate potentialtargets [111–114].The potential benefits of this technology are numerous. Since passive radar systemsnever emit energy (relying only on signals already present), they can operate covertly; targetshave no way of knowing that they are being detected, tracked, and classified by passive