A Clutter Suppression Method Based on SOM-SMOTE Random Forest

Clutter suppression is always a significant but difficult problem in search radars. In complex clutter environment, conventional method may generate a large number of clutter residual plots. Therefore, we propose a novel clutter suppression method based on SOM-SMOTE Random Forest (SOM-SMOTE-RF) from the perspective of classifying targets and clutters. Firstly, we design signal-level features like amplitude, phase, spatial correlation, and range profile of radar echoes in order to construct a high-dimensional feature vector. Secondly, SOM-SMOTE is proposed by combining Self Organizing Map (SOM) and the Synthetic Minority Over-sampling Technique (SMOTE) to solve the imbalanced dataset problem in training Random Forest. Real data shows that SOM-SMOTE-RF suppresses 82% clutters with only 5% targets loss.