Detection Of The BHFSWR First-order Sea Clutter Based On Feature Learning

HF radar will often be interfered by noise, clutter or other pollutant whendetecting targets. As a consequence, false alarm or missed detection will come intobeing in target detection. First-order sea clutter is major in background clutter. Thedetection and tracking of the low velocity targets will be strongly interfered as a resultof it. It also forms a velocity bind zone of detection. The targets in this area can’t bedetected by the feature of sea clutter. Facing this problem, a detection of sea clutter isurgently wanted.This paper starts with the mechanism of the bistatic radar sea clutter for thedetection of sea clutter spectra splitting peaks in complex environment. Firstly, weanalyse the layout of bistatic radar and then deduces the expression of the first-ordersea clutter Doppler shift. We also introduce Gill’s narrow beam cross sections modelsof ocean surface. Based on it, we simulate first-order sea clutter and discuss theinfluence factor of the clutter. After that, we use the model to predict the theoreticalposition of first-order Bragg peaks and analyse the prediction.Based on the prediction of the theoretical position of the Bragg peaks, we presenta detection method of the BHFSWR (Bistatic High Frequency Surface Wave Radar)first-order sea clutter based on features for the detection of sea clutter spectra splittingpeaks. This method searches and detects the Bragg peak candidates by localmaximization, symmetry and continuity detection based on the prior knowledge ofBragg peak theoretical location and the maximal sea current velocity. Further, themethod of splitting peaks detection is proposed. The detection performance of thismethod is verified better than the traditional applying to the measured data.Finally, we propose a detection method of first-order sea clutter based onsupervised learning. this paper introduces an ARTMAP supervision model, and thenwe extract the features of the first-order sea clutter in Range-Doppler spectra. Weclassify each feature and make them numbered forming a feature data base. Then wemake a preliminary analysis on some of feature diagrams. In the end, we put the sortedfeature data along with manual detection data to the ARTMAP supervision model. We can get the predicted positions of the Bragg peaks. After many times of test, we get abest detection result and feature data base. Contrast with the classic method offirst-order sea clutter detection, this method has a high detection rate and the model hasa learning function. It also can greatly avoid false detection with stable algorithm andhas better robustness.