基于递归标记和神经网络的红外目标匹配识别  

Matching and Recogniton of Infrared Target Based on Recursion Marking and Neural Network

作　　者：张恒 

机构地区：[1]哈尔滨工程大学信息与通信工程学院,哈尔滨150001

出　　处：《系统仿真学报》2006年第12期3463-3467,共5页Journal of System Simulation

基　　金：船舶工业国防科技预研基金(05J3.7.2)

摘　　要：提出了一种红外图像匹配算法。选取实际红外目标,依据图像预处理后的图像,提取对应目标的递归标记,作为BP网络的输入样本,目标的形心为输出样本构建BP神经网络。对网络进行抗干扰训练后,根据目标形心特征进行变分辨率相关匹配。该算法在提高了目标识别率的同时,极大地增强了图像处理的实时性,具有工程实用价值。An algorithm for the matching of infrared image was proposed. By selecting the actual infrared target, the recursion marking is assigned to the corresponding target based on image made of segmentation pretreatment. Based on the recursion marking as input sample and the centroid of the target for output, the BP neural network is constructed. After the anti interference training to the network, it carries on correlation matching of alterable resolution by the corresponding centroid characteristics of the infrared target. The algorithm not only improves the recognition rate but also enormously enhances the real-time imagery processing, Moreover the algorithm has the ceaain engineering practicability.

关 键 词：递归标记 神经网络 抗干扰 变分辨率 相关匹配 

分 类 号：TN911.73[电子电信—通信与信息系统]