A Novel Method for Small-Target Detection in Sea Clutter: Spectral Clustering Based on Neighborhood Density Similarity Measure

Small-target detection in sea clutter has always been difficult due to the low signal clutter ratio (SCR). Feature-based detection methods have emerged as a significant area of research in recent years, where pattern recognition methods involve the use of similarity measures. The Gaussian kernel function used in traditional spectral clustering algorithms performs well when handling nonlinear data, however, it may not effectively capture the relationships between sparsely distributed samples. Therefore, this article proposes a novel method for small-target detection in sea clutter: spectral clustering based on neighborhood density similarity measure (SC-ND) to address the aforementioned issues. It incorporates neighborhood density into the Gaussian kernel function and then uses label information to guide the construction of the similarity matrix, at which point a false alarm controllable is achieved. The similarity matrix is then spectrally decomposed to obtain a new representation of the samples, which are subsequently classified using an improved k-means algorithm. Validated on the IPIX 1993 radar dataset, the proposed method achieves a detection probability of 71.1% with an observation time of 1.024 s and a false alarm rate of 0.001. The results are better than fractal-based detector, tri-feature-based detector, and feature-based detector using three TF features.