The unbalanced classification of weak target in sea clutter

The existing classification algorithms for weak targets at sea clutter are difficult to solve the aliasing of single domain features and the imbalance of samples between sea clutter and targets. In this paper, a classifier for weak targets of sea clutter under sample imbalance is studied. First, the features are extracted from multiple domains, including the relative power features of spheres, biplanes and helical scattering from the polarization domain, the relative mean amplitude features from the time domain, and the non-extensive entropy features from the frequency domain. Then the multi-domain features between sea clutter and target are compared and analyzed. Since the number of samples of sea clutter’s features is much larger than that of target’ features, and the sea clutter’s features are local aggregation. In order to solve the classification bias caused by sample imbalance and feature aliasing, a classifier combining K-means and support vector machine (SVM) is designed in this paper. This classifier mainly conducts K-means dynamic clustering of sea clutter samples, divides sea clutter samples originally belonging to one class into multiple classes to alleviate the sample imbalance phenomenon, and then classifies multi-class sea clutter samples and target samples by SVM. The experimental data show that this method has good classification performance.