HRRP Target Recognition Based on Modified Gradient Boosting Decision Tree

Focused on radar automatic target recognition in open set environment, Extreme Value Theory (EVT) is introduced to modify Gradient Boosted Decision Trees (GBDT). We mark the method as Open Set Gradient Boosted Decision Trees (OSGBDT) in the paper. OSGBDT is structured with two steps. First, EVT is utilized to model the distances distribution among samples across features. Positive samples with closely spaced intervals are grouped into the same leaf node during decision tree splitting. Next, EVT is used to model the outputs of prediction models for each class. The probability that a test sample belongs to an unknown class is estimated. Experimental results based on High Resolution Range Profiles (HRRP) database demonstrate that OSGBDT outperforms state-of-the-art open set recognition methods.