mRMR-Tri-ConcaveHull Detector for Floating Small Targets in Sea Clutter
Yanling Shi, Yuefeng Hu
Abstract
Abstract <p xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">For the feature-based detector of small targets in sea clutter, on the one hand, the three-dimensional convex hull-based detector deviates from the distribution of sea clutter vectors in the feature space and only combines the information of low-dimen-sional features. On the other hand, the redundancy and correla-tion between high-dimensional features are high. Consequently, we propose a detector for detecting small floating targets in sea clutter in high-dimensional feature space (HDFS) in this paper. First, the mRMR algorithm to choose the low-relatedness features from the HDFS is proposed. For the mRMR algorithm, we choose the target features and sea clutter features from the 8-dimensional feature space (8-DFS), where the target features and sea clutter features have the highest degree of discrimination in the 3-dimen-sionalfeature space (3-DFS). Second, the distributionof sea clutter in the 3-DFS is concave or convex, which depends on the selection of features. In most cases, the distribution is concave. Using the traditional convex hull to match the concave distribution of sea clutter inevitably enlarges the judgment area and considerably de-creases the detection probability. Due to the concave distribution of sea clutter in the 3-DFS, we propose a new false alarm control-lable three-dimensional concave hull detector based on the mRMR (mRMR-Tri-ConcaveHull detector). In the mRMR-Tri-Concave-Hull detector, the feature vectors in the 3-DFS, which are selected by mRMR, form a concave area that is more suitable for the Con-caveHull detector. Through theexperimentalanalysisof the meas-ured data, we find that the proposed mRMR-Tri-ConcaveHull in this paper can significantly enhance the detection performance compared with the three-feature convex hull detector.