基于标签传播算法的海面漂浮小目标检测方法

在高分辨体制下海杂波与海面小目标具有复杂的特性，特别是对于雷达散射截面积较小的海面漂浮目标，传统的检测方法性能不佳。为了突破临界信杂比情况下的检测性能，可以提取雷达回波的一种或者多种特征，从而进行特征检测，该方法是实现临界信杂比情况下有效检测的重要途经。目前，在3维及以下的特征空间中可以使用凸包学习算法计算判决区域并有效地控制虚警概率，但是在3维以上的特征空间中凸包学习算法计算复杂度提高，难以进行检测。针对这个问题，该文提出一种基于标签传播算法的海面小目标检测方法，它突破了凸包学习算法的维数限制和决策域必须为凸集的形状限制，能够在高维特征空间进行检测并有效地控制虚警。经过实测数据集验证，基于标签传播算法的海面小目标检测方法在0.512 s和1.024 s的观测时间内分别获得了88.4%和92.0%的检测概率，相比于基于K近邻(KNN)的检测器有了3.3%和2.8%的检测概率提升。

Small Target Detection on Sea Surface Based on Label Propagation Algorithm

Sea clutter and small targets have complex characteristics in the high-resolution radar system. For the target with small radar cross section, the traditional detection method has limited detection performance. In order to break through the critical signal to clutter ratio state, one or more features of radar echo can be extracted for joint feature detection, which is an important way to achieve effective detection in the case of critical signal to clutter ratio. At present, convex hull learning algorithm can be used to calculate the decision region and control effectively the false alarm probability in the feature space of three dimensions and below, but the computational complexity of convex hull learning algorithm is increased above the feature space, and make it difficult to detect target. To solve this problem, a small target detection method based on label propagation algorithm is proposed. It can detect small target in high-dimensional feature space and the false alarm can be effectively controlled. The experimental results on the actual database show, the detection probabilities of 88.4% and 92.0% are obtained in 0.512 s and 1.024 s respectively, which are 3.3% and 2.8% higher than those of the K-Nearest Neighbor (KNN) detector.