Modern RF Radar signal processing has been receiving much attention for wide range of domains that include industrial, environmental, and military applications.Inherently, the received raw spatial-temporal signals can be 1-D, 2-D, or 3-D and are usually of uncertain nature, because of changing conditions and optical background variations. In this paper, we apply novel concepts for hyper-neural theory that allow for incorporation of variables attribute definitions and uncertainties for the purpose of effective evidential learning and subsequent key output features determination in the radar processing. Application to wide-band angle of arrival data sets at several carrier frequencies has been carried out in order to illustrate the strengths as well weakness of the approach. Using interval set-based operations together with segmentation of the data is proved useful and gave good sensitivity of detection.