Detection method based on nonlinear feature re-expression with variance maximization

Detecting small maritime targets against sea clutter using radar poses significant challenges. Multidimensional feature based detection methods have proven effective, but current approaches often rely on linear correlations among features, which do not reflect real-world scenarios accurately. To address this, we first analyze the correlations among features in various measured scenarios. We then propose a target detection method in sea clutter that employs nonlinear feature re-expression based on variance maximization. This method fits the nonlinear relationships between features to identify the optimal projection direction, achieving variance-optimal nonlinear projection. This results in new features with enhanced classification capability for distinguishing between sea clutter and targets. Testing with measured data from the SDRDSP public dataset demonstrates that the proposed detection method outperforms existing methods based on linear feature re-expression in terms of detection performance.