Automatic target recognition(ART) using Synthetic Aperture Radar(SAR) imagery is investigated in this paper. The feature extraction problem of SAR imagery is first analyzed

then a matching scheme that incorporates relative distance and magnitude between features is investigated. Following the Birkhoff-von Neumann theorem

we relax the match matrix constraints from permutation matrix constraints to doubly stochastic matrix constraints. Via Lagrange multipliers and a barrier function

the constraints are incorporated into the objective function

and the matching problem is posed as a nonlinear optimization problem. Using a combination of deterministic annealing and softassign

the objective function describing the matching problem is minimized. Recognition is performed by comparing the costs of the matches between the test image and deferent pattens. To account for the difference in the number of features

the computed costs are first caled by the ratio of the number of features between the two images. The test costs image belongs to the class of the pattern with the smallest acaled matching cost. Experimental results show the power of this approach in SAR target recognition.