A W-KNN classifier to improve radar outlier rejection performance

Radar automatic target recognition (ATR) mainly corresponds to uncooperative targets, and the training database is usually incomplete. So in the test step, we should reject the targets with new class labels as outliers firstly, and then recognize the remaining targets (inners) in detail. Combining with engineering application, we proposed a reasonable method to artificially generate outliers and designed a weighted KNN (W-KNN) classifier to treat with the outlier rejection problem. Experiments conducted on high-resolution range profiles (HRRP) data show that the W-KNN classifier is a promisingly method to treat with the rejection problem.