多DSP并行的神经网络集成目标识别法  

Neural Network Ensemble Target Recognize of Parallel Structure based on DSP

作　　者：鄢余武[1,2] 胡方方[1] 张金成[1] 

机构地区：[1]空军工程大学导弹学院,陕西三原713800 [2]解放军理工大学军事网格研究中心,江苏南京210007

出　　处：《火力与指挥控制》2009年第9期99-102,共4页Fire Control & Command Control

基　　金：空工大学术联合基金资助项目

摘　　要：针对分布式信息融合结构、异质传感器条件下的目标识别问题,提出了基于多DSP并行结构的神经网络集成目标识别方法。给出了生成神经网络集成的具体方法,并构造了一个实际的空中目标识别硬件系统。结果表明,系统的目标识别性能明显优于单个神经网络的目标识别性能,且识别目标的速度很快。Aiming at the problem in target recognize when multi-sensor are distributed, a method for target recognizes based on neural network ensemble and parallel structure of multiple DSP was proposed. Introduced a technique to ensemble several neural networks, then constructed a real target recognize system based on the method which is proposed prior. Simulation results indicated that the performance of the real target recognize system which use the neural network ensemble is better than the older ones which without neural network ensemble, and the speed is very fast.

关 键 词：DSP 并行 目标识别 神经网络集成 

分 类 号：E839[军事—战术学] TP274[自动化与计算机技术—检测技术与自动化装置]