SAR Target Discrimination based on the Ensemble Projection Feature

The traditional discrimination features for SAR target discrimination are usually extracted in an unsupervised manner, leading to the fact that the description ability of discrimination features are insufficient and the discrimination performance degrades in complex scenes. Aiming at this problem, this paper proposes a supervised SAR target discrimination feature based on ensemble projection. Firstly, the sample prototype sets are automatically sampled from the training samples by using the Max-Min sampling method to represent a series of visual attributes. Secondly, the discrimination function is learned based on each prototype of the sample prototype sets. Finally, the samples are projected to the prototype functions and the projection values are concatenated as the final discrimination features. We conduct the target discrimination experiments on the measured data of miniSAR and FARAD SAR, and the results demonstrate that compared with the traditional discriminative features, the proposed discrimination feature can achieve better discriminant performance.