Small Floating Target Detection on Sea Surface Using Label Propagation Algorithm With High-Dimensional Graph Features

Detecting small targets in complex marine environments is challenging for radar systems. Existing methods relying on high-dimensional features often fail to account for correlations in radar echo data and do not fully capture the structural differences between sea clutter and target echoes, limiting radar detection performance. To that end, this paper proposes a novel detection approach for small marine targets utilizing a label propagation algorithm that incorporates high-dimensional graph features. First, a visibility graph topology is constructed to model the correlation between the phase entries of radar echo. Then, building upon the constructed visibility graph model, three quantitative graph features are defined, including the visibility graph entropy, the maximum eigenvalue of the Laplacian matrix, and the number of fully connected nodes. By combining these with other three available graph features, the detection of small sea surface targets is transformed into an anomaly detection problem in a six-dimensional measurement feature space. Finally, a detector is constructed by combining the high-dimensional graph features with an improved label propagation algorithm, enabling effective detection of targets in the defined measurement feature space. Experimental results and comparisons with existing detection methods have been included to demonstrate the effectiveness of the proposed detector in enhancing target detection performance.