Feature Selection Method for Open Set HRRP Recognition Based on Local Outlier Factor

‍ ‍Feature selection is a crucial part of the radar target recognition process. Recognition accuracy and efficiency can be enhanced by selecting high-quality features from the original set of features to form a new subset. In practical applications， the existing feature selection methods， which are based on the closed-set assumption， may cause performance degradation in open set recognition （OSR） caused by the presence of out-of-database targets. To enhance the recognition performance of high-resolution range profiles （HRRPs） in open environments， a feature selection method for HRRP OSR based on the local outlier factor （LOF） is proposed in this study. First， 15-dimensional feature vectors are extracted from the original HRRPs as the initial feature set. Second， the concept of feature aggregation is introduced and the LOF is used for its measurement， ensuring minimal open space risk during OSR by evaluating the aggregation of feature subsets. Additionally， the separability of feature subsets is assessed using the centroid method to guarantee good inter-class discrimination. Finally， the feature selection process is optimized using a forward search algorithm to ensure that the selected feature subset is the optimal solution under dimensional constraints. Experimental results demonstrate that the proposed method outperforms mainstream feature extraction methods regarding features selected in open set environments， and the recognition performance of HRRPs in OSR scenarios is improved.