Decision-Level Fusion Performance Improvement From Enhanced HRR Radar Clutter Suppression

Many surveillance systems incorporate High Range Resolution (HRR) radar and Synthetic Aperture Radar (SAR) modes to be able to capture moving and stationary targets. Feature-, signature-, and categorical-aided tracking and Automatic Target Recognition (ATR) applications benefit from HRR radar processing. Successful Simultaneous Tracking and Identification (STID) [6, 12, 65] applications exploit feature information to determine the target type and dynamics. Throughout the paper, we use identification, as opposed to recognition, to clarify the process of distinguishing between two targets of the same classification label or allegiance type. To maximize a search area, airborne systems operate at standoff ranges to detect targets and initiate tracks [3, 5]. After target acquisition and track initiation, tracking systems then transition into a track maintenance mode. However, closely spaced targets require feature analysis to identify the targets. In track maintenance, HRR radar affords dynamic processing analysis for vehicle tracking and signal feature extraction (range, angle, aspect, and peak amplitudes) for target detection and identification [7]. Pattern recognition algorithms applied to ATR problems are typically trained on a group of desired objects in a library to gain a statistical representation of each objects’ features. One-dimensional (1-D) HRR classifiers exploit the location and peak amplitude information contained in the HRR signatures [19, 38]. HRR classifiers align input signatures to the library templates or models [16] and determine the best correlation value for the aligned features. HRR ATR algorithms often apply a threshold to the best score to reject questionable objects before making identification or class label decisions. As per the previous work on target identification from HRR signatures, we improve existing capabilities by increasing the peak amplitudes and refine range-bin locations through clutter suppression techniques. A number of papers have been published that evaluate one-dimensional (1-D) HRR ATR solutions [22, 27, 46, 62, 63]. Classifiers have been developed for correlation [34], Bayes and Dempster Shafer information fusion [11], entropy and information theory analysis [8], and neuro-fuzzy methods [10]. The classifier results have been used for tracking [9, 66] and multi-look HRR identification [53, 67]. Other approaches include eigenvalue template matching [51], Eigen-Extended Maximum Average Correlation (EEMACH) filters [31] and likelihood methods accounting for Rician, amplitude, specular, and diffuse, Cisoid scattering [18]. Since we utilize a combination of sensor and exploitation algorithms (with reported decisions) we are not afforded feature or signal-level fusion options. Using inspiration from the above ATR fusion methods, we incorporate maximum-likelihood Bayesian methods into our novel Confusion Matrix Decision-Level Fusion (CM-DLF) algorithm.