Analysis of boundary performance in feature time series detection in marine environments

Feature detection within sea clutter environments is recognized for its rapid computation, high separability, and stability. By leveraging the temporal attributes of features and integrating historical frame data with current frame features, it is possible to further enhance feature separability, thereby optimizing detector performance. Real-world radar target detection scenarios are often fraught with challenges such as high sea states, low signal-to-noise ratios, and limited pulse accumulations. Understanding the boundary performance of feature temporal series detection under these adverse conditions is critical for effective maritime target detection. This paper employs an Autoregressive (AR) model for feature modeling and one-step prediction of radar echo data. The predicted features are fused with observed features to derive composite features, which are then subjected to convex hull detection. Using a dataset provided by the Naval Aviation University, this study conducts experiments to detect actual buoy measurements at sea. Furthermore, simulated targets are introduced to comprehensively analyze the boundary performance of the temporal series detector, thereby providing valuable references for the application of feature temporal series detectors.