Small Target Classification in Sea Clutter Using Graph-based Maximum Connected-component Learning Algorithm via Alternative Shrinkage with Rejection Ability

Xiaohui Bai

doi:10.1016/j.dsp.2026.106345

Small Target Classification in Sea Clutter Using Graph-based Maximum Connected-component Learning Algorithm via Alternative Shrinkage with Rejection Ability

2026-06-15

0

PRE

AI

DOI：10.1016/j.dsp.2026.106345

Original

Original request for help

Save

Abstract

En 中文

Accurate classification of small targets within sea clutter remains a significant challenge in radar signal processing. Existing methods suffer from limited scalability to high-dimensional features and an inability to reject unknown targets. This paper proposes a novel classification framework based on a maximum connected-component learning algorithm via alternative shrinkage, which constructs non-overlapping decision regions under connectivity constraints, adapting effectively to non-convex sample distributions. Seven discriminative features are extracted from the amplitude, Doppler, and time-frequency domains to form the feature space. An alternating shrinkage strategy resolves decision-region overlaps while preserving balanced classification probabilities across classes. Experiments on the IPIX radar database show that the proposed method outperforms SVM, decision tree, and convexhull‑based classifiers. It achieves high probability of correct classification for known targets and high probability of correct rejection for unknown targets. The proposed framework offers a robust and scalable solution for sea-surface target classification, with potential applicability in other domains.

Small Target Classification in Sea Clutter Using Graph-based Maximum Connected-component Learning Algorithm via Alternative Shrinkage with Rejection Ability

Abstract

Journal

Organization