Detection of Small Targets on the Sea Surface Based on High-Dimensional ConvH Classifier

This article proposes a viable and easy-to-operate one-class classifier (OCC) of high-dimensional ConvH (high-D ConvH), under the background of feature detection for small targets on the sea surface. Our classifier breaks the bottleneck of current ConvH classifier, i.e., being confined only to the detection within 3-D feature space, by addressing the following two issues: 1) the current ConvH decision criterion is limited to 3-D and 2) the unacceptable learning time of ConvH under high-D space. For the first issue, derived from the half-space description of ConvH, we propose a fast ConvH decision criterion, which can be extended to any dimensional space and owns lower algorithmic complexity. Accordingly, a corresponding fast ConvH decision algorithm is designed. For the second issue, we first give an approach to accelerate ConvH learning, based on which we propose two fast ConvH learning algorithms of different mechanisms. The rigorous mathematical derivations prove that the proposed algorithms are correct and feasible. Combining the above algorithms, the high-D ConvH classifier is obtained, and then a novel feature detection method for small target on the sea surface is presented by integrating the high-D ConvH classifier into existing feature detection process. The experiments on the CSIR datasets reveal that: 1) the high-D ConvH classifier is successfully applied to the 7-D feature space, and it owns shorter learning and decision time than existing methods and 2) the proposed detection method owns superior detection performance compared with its competitors in high-D space.