# Mechanism–Data Collaboration for Characterizing Sea Clutter Properties and Training Sample Selection

## TL;DR

This paper introduces a new method to improve radar detection in maritime environments by better modeling sea clutter and selecting training samples.

## Contribution

The novel approach combines mechanism and data to model sea clutter and optimize training samples for radar detection.

## Key findings

- The proposed method accurately characterizes sea clutter properties using scattering coefficients and fusion strategies.
- Hybrid feature selection improves training sample diversity and compactness, enhancing detection performance.
- Detectors trained with the proposed samples show strong generalization across different maritime environments.

## Abstract

Multi-feature-based maritime radar target detection algorithms often rely on statistical models to accurately characterize sea clutter variations. However, it is a big challenge for these models to accurately characterize sea clutter due to the complexity of the marine environment. Moreover, the distribution of training samples captured from dynamic observation conditions is imbalanced. These multi-features extracted from inaccurate models and imbalanced data lead to overfitting or underfitting and degrade detection performance. To tackle these challenges, this paper proposes a mechanism–data collaborative method using the scattering coefficient as a representative feature. By establishing a mapping relationship between measured data and empirical values, the classical model is piecewise fitted to the measured data. A fusion strategy is then used to compensate for interval discontinuities, enabling accurate characterization of clutter properties in the current maritime environment. Based on the characterized clutter properties, a hybrid feature selection strategy is further proposed to construct a diverse and compact training sample set by integrating global density distribution with local gradient variation. The experiments based on field data are included to evaluate the effectiveness of the proposed method including sea clutter characterization accuracy and training sample selection across various scenarios. Experimental results demonstrate that the proposed method provides a more accurate representation of sea clutter characteristics. Moreover, the detectors trained with the proposed training samples exhibit strong generalization capability across diverse maritime environments under the condition of identical features and classifiers. These achievements highlight the importance of accurate sea clutter modeling and optimal training sample selection in improving target detection performance and ensuring the reliability of radar-based maritime surveillance.