多次观测信号检测的最优量化  

Optimum Quantization for Multi-Observation Signal Detection

作　　者：王瑛[1] 邹芳[2] 王飞雪[1] 孙广富[1] 

机构地区：[1]国防科技大学电子科学与工程学院,湖南长沙410073 [2]西安武警工程学院通信工程系,陕西西安710086

出　　处：《电子学报》2008年第3期575-580,共6页Acta Electronica Sinica

基　　金：新世纪优秀人才支持计划(No.NCET-04-0995)

摘　　要：信号检测的最优量化是降低实现复杂度的重要手段．在有限精度量化前提下，本文推导了多次观测信号检测的检测性能（检测概率和虚警概率），并证明了实际信号环境中多次观测积累值是最佳统计量，文中进一步提出了基于遗传算法的最优量化设计方法，可方便地优化等效检测能力因子等各种检测性能指标，离线设计最优量化参数，以红外弱小目标检测多帧图像预积累为例，最优量化设计可将量化比特数由通常的8～14比特减少到2～4比特，有效降低图像处理实现复杂度。Optimum quanfization for signal detection can effectively reduce the complexity of realization. Considering finite precision quantization, we analyze the performance of multi-observation signal detection,and prove that multi-observation accumulation is the optimum statistic in real-world signal environments. Furthermore, we propose a novel approach based on genetic algorithm to design optimum quantizers offline, which optimize figures of merit for signal detection such as the equivalent detectability factor. The approach is applied to the design of multi-frame accumulation for infrared dim target detection. Numerical results show that optimum quantization can cut down the conventional 8-14bitsquantization to 2-4bits, thus effectively reduce the complexity of image processing.

关 键 词：信号检测 最优量化 积累 遗传算法 

分 类 号：TN911.23[电子电信—通信与信息系统]