摘要
该文针对二元假设检验问题,首先在贝叶斯准则的基础上,分析了最小化贝叶斯代价所对应的最优噪声,将贝叶斯代价的最小化问题等价为虚警概率和/或检测概率的最优化。其次,在保证一定虚警概率和检测概率的前提下,建立起能同时改善检测概率和虚警概率的模型。然后分别给出当检测概率一定时虚警概率最小和虚警概率一定时检测概率最大这两种极限情况下对应的最优加性噪声,并对其进行线性凸组合以获得模型所需的最优加性噪声,进一步分析并证明了该模型能够成立的充分条件。再次,获得先验概率已知和未知两种情况下最小化贝叶斯代价时所对应的加性噪声,且当先验知识发生改变时,该算法只需调整加性噪声中一个可变参数即可获得相应的最优贝叶斯代价。最后,结合具体的检测问题,通过仿真验证了所提算法的有效性。
The optimal noise that minimizes Bayes risk for a binary hypothesis testing problem is analyzed firstly. As a result, the minimization of Bayes risk can be equivalent as the optimization of the detection probability (PD) and/or false alarm probability (PFA) . Secondly, a noise enhanced model, which can increase PD and decrease PFA simultaneously, is established under the premise of maintaining predefined PFA and PD. Then the optimal additional noise of this model is obtained by a convex combination of two optimal noises of two limit cases, which are the minimization of PFA with maintaining the predefined PD and the maximization of PD with maintaining the predefined PFA, respectively. Furthermore, the sufficient conditions for this model are given. What's more, the additive noise that minimizes the Bayes risk is determined when the prior probabilities are known or not, and the corresponding additive noise can be obtained by recalculating a parameter only if the prior information changes. Finally, the availability of algorithm is proved through the simulation combined with a specific detection example.
作者
刘书君
杨婷
唐明春
王品
李勇明
LIU Shujun YANG Ting TANG Mingchun WANG Pin LI Yongming(College of Communication Engineering, Chongqing University, Chongqing 400044, Chin)
出处
《电子与信息学报》
EI
CSCD
北大核心
2017年第2期293-300,共8页
Journal of Electronics & Information Technology
基金
重庆市基础与前沿研究(cstc2016jcyj A0134
cstc2016 jcyj A0043)
国家自然科学基金(61501072
61301224
41404027
61108086
61471072)
重庆市社会事业与民生保障专项(cstc2016 shmszx40002)
中央高校重点基金(CDJZR155507)~~
关键词
信号处理
贝叶斯准则
噪声增强模型
加性噪声
假设检验
Signal processing
Bayes criterion
Noise enhanced model
Additive noise
Hypothesis testing
作者简介
通信作者:刘书君:女,1981年生,讲师,研究方向为统计信号分析、图像处理.liusj@cqu.edu.cn
杨婷:女,1990年生,博士生,研究方向为噪声增强信号处理、统计信号分析.
唐明春:男,1981年生,研究员,研究方向为天线与电波传播、微波毫米波理论与技术.
王品:女,1979年生,副教授,研究方向为智能信息处理、光学信息处理.
李勇明:男,1976年生,副教授,研究方向为智能计算与模式识别、数据挖掘.