Detection and Mitigation of Ground Clutter in Polarimetric Phased Array Radar Measurements Using Machine Learning and Physics-Based Discriminants

This article presents clutter detection and mitigation for polarimetric phased array weather radar measurements using machine learning. The following three approaches are analyzed for clutter detection in the cylindrical polarimetric phased array radar (PAR) measurements, including naive Bayes classifier (NBC), multilayer perceptron (MLP), and convolutional neural network (CNN). Results show that CNN achieves the best performance in clutter detection, followed by MLP and NBC. This is because CNN utilizes spatial information of the input images, which has different features for clutter from that for weather. It is also shown that the combination of physics-based discriminants of power ratio and raw radar measurements is more effective in clutter detection than the direct use of raw radar measurements. In addition, CNN is employed for clutter mitigation and its performance is compared with the traditional speckle filter technique. It is demonstrated that CNN outperforms the speckle filter and incorporation of power ratio in the training process could further improve CNN’s performance in clutter mitigation.