一种基于置信函数的分类器自优化雷达点迹识别算法

雷达组网协同探测中，受不同探测精度、观测维度及环境噪声影响，信息系统获取的传感数据包含一定不精确、不确定信息，导致无法对目标点迹准确分类识别。为此提出了一种基于置信函数的分类器自优化雷达点迹识别算法。首先，基于置信函数理论创建目标、杂波、不确定数据的证据识别框架，并设计可实时给定目标数据类别隶属度的深度神经网络模型分类器。然后，依托当前迭代轮次分类结果进行辅助决策证据构建，并根据点迹分布特性进行证据修正融合。最后，基于全局融合结果进行点迹类别标签更新，并重新驱动网络模型分类器进行在线学习与更新，如此迭代循环直至所有的雷达点迹数据类别标签不再发生改变。基于雷达实测数据集对算法性能进行验证分析，结果表明与传统算法相比新算法能够有效提升雷达点迹的分类正确率，而且随着样本数据的丰富算法收敛时间可急速减少，便于在后续工程中推广应用。

A Recognition Algorithm of Radar Plots Based on Confidence Function and Self-updating Classifier

In the process of radar networking and cooperative detection, affected by different detection accuracy, observation dimension and environmental noise, the sensor data obtained by the information system contains certain imprecise and uncertain information, which will result in that the target points cannot be accurately classified and identified. In order to solve this problem, a recognition algorithm of radar plots based on confidence function and self-updating classifier (RARP-CFSC) is proposed in this paper. First, the evidence recognition framework based on the confidence function theory for target, clutter and uncertain data is created, and a deep neural network model classifier that can give the membership of target data category is also designed. Then, according to the classification results of the current iteration round, the decision support evidence set is constructed, and the modified fusion is also carried out according to the characteristics of the plot distribution. Finally, based on the global fusion results, the plot category labels are updated, and the network model classifier is re trained and updated online, so as to iterate until all the radar plot data category labels are no longer changed. The performance of the algorithm is verified and analyzed based on the real radar measured data set. The results show that the new algorithm can effectively improve the classification accuracy of radar plots compared with the traditional algorithm, and the algorithm convergence time can be rapidly reduced with the enrichment of sample data, which is convenient for promotion and application in subsequent projects.