Abstract:

It is still an open issue how to deal with the effect of measurement scale on features in the domain of full-polarized high range resolution profile (HRRP) feature extraction based on H/A/α decomposition. To solve it, the feature subsets are extracted under the mean measurement scale. Meanwhile, the joint dynamic mutual information is introduced to select the optimal mean measurement scale and exclude the redundant features in the subsets. Finally, a novel target recognition approach based on full-polarized HRRP with the combination of Bagging and Boosting comes into being. The experiments on multi-class plane targets validate the proposed method’s efficiency.