High-Resolution Radar Clutter Suppression Method on Riemannian Manifold Based on Intelligent Semantic Segmentation in Complicated Environments

‍ ‍Radar detection of low-altitude targets， such as migrating biological entities and unmanned aerial vehicles （UAVs）， is heavily impacted by complex clutter environments， resulting in increased missed detections and false alarms. Addressing this challenge requires effective clutter suppression methods. This paper introduces a clutter suppression method based on intelligent semantic segmentation on the Riemannian manifold， designed to handle high-resolution clutter in complicated environments. In such scenarios， low-altitude weather clutter and ground clutter are superimposed， with clutter types and characteristics exhibiting spatial variability. The proposed method employs intelligent semantic segmentation using a U-net architecture to distinguish various clutter types in the range-Doppler domain. This segmentation identifies clutter regions and informs subsequent clutter suppression strategies. A clutter semantic segmentation dataset， constructed using real rain and ground clutter data collected by a high-resolution phased array radar， was used to train the intelligent segmentation network. The semantic segmentation process categorizes high-resolution radar echoes into subbands in the Doppler dimension based on clutter type. For each clutter region， the clutter power spectral density （PSD） is estimated using the geometric mean of reference cell samples on the Riemannian manifold. Reference cells are selected based on segmentation results， ensuring that samples with consistent time-frequency characteristics are used for estimation. Finally， an adaptive clutter suppression filter is constructed using the clutter PSD to achieve effective suppression. The proposed method is validated using real high-resolution clutter data obtained under diverse weather conditions and radar positions. Experimental results from two scenarios—real clutter data with simulated targets and real data containing both clutter and targets—demonstrate that the method effectively suppresses clutter in complex， multi-type environments. This approach significantly enhances the signal-to-clutter ratio of targets， improving detection performance in challenging conditions.