粗糙K-means和AdaBoost结合的雷达辐射源快速识别算法

A Fast Radar Emitter Recognition Algorithm Based on Rough K-means Combined with AdaBoost

针对数据集识别难度分布不均匀,提出了基于粗糙K-means和AdaBoost的雷达辐射源快速识别算法。该算法由2个阶段构成:初级识别阶段提出一种改进粗糙K-means算法,将数据特征空间分割为确定区域、粗糙区域和不确定区域,构建雷达辐射源快速识别算法模型,对数据集进行筛选和识别,同时提出了一种确定粗糙K-means算法初始聚类中心和聚类数量以解决其固有缺陷的思路;在高级识别阶段,基于粗糙区域已知数据训练的多类AdaBoost分类器识别不确定区域未知数据,提升识别精度。仿真结果表明:该算法与RBF-SVM和AdaBoost相比,精度浮动在-0.1%到+1.4%之间,训练时间和测试时间分别最大缩短0.857s和0.005s,在保持了较高识别精度和泛化能力的同时,明显降低了计算复杂度,缩短了耗时,提供了设计雷达辐射源快速识别算法的新思路。

Aimed at the fact that data samples in the same data set are difficult to recognize because of mal- distribution, this paper proposes a fast radar emitter recognition algorithm based on rough K-means com- bined with AdaBoost. The algorithm is composed of two stages. At the primary recognition stage, an improved rough k-means algorithm is proposed, and the data feature space is divided into the certain area, the rough area and the uncertain area to construct a fast radar emitter recognition algorithm model so as to filter and recognize the data set. And at the same time a heuristic approach is proposed to solve the inherent shortcomings of the original rough K-means by ascertaining its initial clustering number and centers. And at the advanced recognition stage, unknown samples dwelling in the uncertain area are recognized by the multi-class AdaBoost classifier trained by the unknown ones in the rough area, thus promoting the recog- nition accuracy of the algorithm. The simulation results show that compared to RBF-SVM and AdaBoost, the scope of an accuracy fluctuation is from -0.1% to ＋1.4%, the shrinkage of a training time is 0.857 s,and the shrinkage of a test time is 0.005 s at most, and apparently the computational complexity is lowered and the time consumed is shortened respectively by using this new algorithm under conditions of reserving comparatively high recognition accuracy and generalization capability. By so doing, this provides fast radar emitter recognition algorithms-designing with new train of thought.