A Novel Approach of Automatic Target Recognition using High Resolution Radar Range Profiles

Generalized information cut (GIC) based on the nonlinear feature mapping using the kernel function is invariant for the linear transformation and the coordinates shift. By comparison and analysis of the angle similarity measurement and generalized information cut, the criterion of generalized information cut is proposed to solve the automatic target recognition using high resolution radar range profile. Feature mapping and inner product is synthesized in the feature space. The range profile data established from four different scaled aircraft models in unclassifiable low dimensional space is mapped into classifiable high dimensional space via the kernel function, and then the linear classifiable pattern transformed from nonlinear unclassifiable pattern is accomplished. Experimental results suggest that the target recognition technique based on the proposed method achieves better anti-noise performance, and is much more efficient in the improvement of recognition ratio of four different aircrafts than using the conventional angle similarity coefficient (the vectorial angle cosine).