基于压缩感知的^(252)Cf源驱动核材料浓度识别技术研究  

^(252)Cf-source-driven nuclear material concentration identification based on compressive sensing

作　　者：李鹏程[1] 魏彪[1] 冯鹏[1] 何鹏[1] 周密[1] 米德伶[1] 任勇[2] 

机构地区：[1]重庆大学光电技术及系统教育部重点实验室,重庆400044 [2]重庆大学通信工程学院,重庆400044

出　　处：《强激光与粒子束》2015年第7期172-177,共6页High Power Laser and Particle Beams

基　　金：国家自然科学基金项目(61175005);中央高校基本科研业务费专项资金项目(CDJXS11121145)

摘　　要：针对252Cf源驱动噪声分析测量法中核材料浓度识别问题,采用压缩感知理论,在K最近邻(KNN)识别算法基础上,研究了一种基于压缩采样的K最近邻(CSKNN)分类识别方法,进而研究并分析了CSKNN方法的识别概率。实验结果表明,CSKNN分类识别方法只需少量的观测值(观测比M/N≥0.1),即可达到分类识别的目的;当信噪比提高时,识别概率将会以更快的速度收敛至100%,且对K值的敏感程度也会随之降低。这样,不仅提高了核军控核查的实时性,而且还有效降低了采样成本,为核材料浓度的在线判读提供了一种新的理论基础和实现方法。For solving the identification problem in 252 Cf source driven noise analysis method, we used the compressive sens ing theory and the nearest neighbor recognition algorithm, proposed a new classification method named CSKNN method, and then analysed identification probability. The experimental results show that for the classification and identification purposes, the CSKNN identification method only needs a few observations （the ratio between the number of measured values and the fission neutron signal length is no less than 0.1）. When the signal to noise ratio increases, the recognition probability will converge faster to 100% and be less sensitive to K. Hence, the CSKNN method is reasonable and feasible, not only because it improves the real time performance of nuclear arms control verification, but also effectively reduces the sampling cost. Most importantly, it pro vides a new theoretical basis and implementation method for the online classification of nuclear material concentration.

关 键 词：252Cf源噪声分析法 压缩采样 观测比 K最近邻识别算法 识别概率 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]