基于半监督学习聚类数据标注的多功能雷达工作模式识别

Multi-function radar work mode recognition based on clustering data annotation

摘要:

对多功能相控阵雷达工作模式准确识别可为电子对抗决策提供依据，具有重要研究意义。现有工作模式识别方法主要基于已知标签的训练集，而实际中先验信息匮乏，数据标签难以获知，极大影响了工作模式识别性能。为此，提出了一种在少量先验信息辅助下基于半监督学习聚类实现未知数据标注的工作模式识别方法。首先根据聚类算法的内部评价指标和外部评价指标对比分析AP聚类(affinity propagation clustering)、DBSCAN聚类(density-based spatial clustering of applications with noise)和模糊C均值聚类(fuzzy C-means clustering, FCM)3种典型聚类算法的性能，验证了AP聚类算法性能最优，并将其应用于对截获数据的数据标注中。然后利用卷积神经网络对雷达工作模式进行识别，并与已知标签训练集下的网络进行对比，验证了基于AP聚类算法进行数据标注的可行性，提升了相较传统识别网络的抗噪性，为后续多功能雷达行为认知提供了基础。

Abstract:

The accurate identification of the working mode of multi-function phased array radar can provide a basis for electronic countermeasure decision-making, which is of great research significance. The existing working pattern recognition methods are mainly based on the training set of known labels, but in practice, there is a lack of prior information, and the data labels are difficult to obtain, which greatly affects the performance of working pattern recognition. In this paper, a working pattern recognition method based on semi-supervised learning to achieve unknown data labeling with the help of a small amount of prior information is proposed. Firstly, the performance of AP clustering(affinity propagation clustering), DBSCAN(density-based spatial clustering of applications with noise) and FCM(fuzzy C-means clustering, FCM) is compared and analyzed according to the internal evaluation index and external evaluation index of the clustering algorithm, and it is applied to the data annotation of the intercepted data after the optimal performance of the AP clustering algorithm is verified. Then, the convolutional neural network is used to identify the radar working mode and compare it with the network under the known label training set, which verifies the feasibility of data annotation based on AP clustering algorithm, improves the noise immunity compared with the traditional recognition network, and provides a basis for subsequent multi-functional radar behavior cognition.