非合作目标样本极端缺失下的辐射源个体识别方法研究

Specific emitter recognition is one of the research hotspots and key technologies in the field of electronic countermeasure, which is considered as an important foundation to improve our country electronic defense ability. However, in practical applications, the samples of non-cooperative targets with high threat degree are hard to obtain. In this case, the existing classifiers cannot achieve satisfactory recognition performance, which has become a bottleneck for the development of specific emitter recognition system. For this problem, this project will study on intelligent classification method, and focus on the following aspects: (1) studying a novel classification criterion based on minimizing the overall misclassification threat degree, and establishing a corresponding evaluating index based on false alarm rate and the overall misclassification threat degree; (2) Based on the above classification criterion, studying a multiclass data description method to solve extreme imbalance learning problem; (3) studying an online data description method based on key information updating, which can achieve real-time upgrading without the large-scale database. This project aims at proposing an intelligent classification method suited to the demand of specific emitter recognition, and the achievements can be easily extended to most of the automatic recognition systems, which will provide the corresponding theoretical and technical supports to enhance the performance of our electronic countermeasure system.

辐射源个体识别是电子对抗领域的研究热点和亟需解决的关键技术之一，是提高我国电子防御能力和水平的重要保障。然而，实际应用中，辐射源识别面临具有高威胁度的非合作目标样本极端缺失的问题，致使现有分类器性能下降甚至完全失效，这已成为制约辐射源个体识别系统发展的一个瓶颈。本项目针对该问题，开展智能分类方法研究，主要内容包括：（1）研究基于总体威胁最小化的判决准则，并建立与之配套的综合虚警率和总体误分威胁度的评价指标；（2）以上述判决准则为基础，研究针对样本极端不平衡问题的多类别数据描述方法；（3）研究基于关键信息更新的在线学习数据描述，使得系统可以在不依赖大规模数据库的基础上实时进行升级。本项目旨在构建一套符合辐射源个体识别实际需求的智能分类方法，该研究成果可以较容易地推广至大多数自动识别系统，从而为我国电子对抗体系整体性能的提升提供理论与技术支撑。

辐射源个体识别是电子对抗领域的研究热点和亟需解决的关键技术之一，是提高我国电子防御能力和水平的重要保障。然而，实际应用中，辐射源识别面临具有高威胁度的非合作目标样本极端缺失的问题，致使现有分类器性能下降甚至完全失效，这已成为制约辐射源个体识别系统发展的一个瓶颈。本项目针对该问题，开展智能分类方法研究，主要内容包括：（1）研究基于总体威胁最小化的判决准则，并建立与之配套的综合虚警率和总体误分威胁度的评价指标；（2）以上述判决准则为基础，研究针对样本极端不平衡问题的多类别数据描述方法；（3）研究基于关键信息更新的在线学习数据描述，使得系统可以在不依赖大规模数据库的基础上实时进行升级。