图像特征的伪装效果评估技术  

Camouflage effectiveness evaluation based on image feature

作　　者：崔宝生 薛士强 姬艳军 于宏宇 张勇 

机构地区：[1]总装备部工程兵科研一所,江苏无锡214035

出　　处：《红外与激光工程》2010年第6期1178-1183,共6页Infrared and Laser Engineering

摘　　要：定义以目标形心为中心、以目标图像的轮廓为模板的八联通区域作为该目标伪装效果评估的直接背景。依据目标识别机理和相关的伪装原则,从图像的统计、形状和纹理三个方面分别筛选并提取出目标及其8个直接背景的19个特征值;根据3σ法则,对提取的特征值进行归一化处理后构成9行19列评估矩阵,在此基础上,采用BP神经网络建立了伪装效果量化评估模型。利用积累的大量工程伪装检测数据开展了系统试验;利用获取的试验数据作为样本集,对模型进行了训练、验证和测试。试验结果表明:所建立的伪装效果量化评估模型预测结果与专家评估结果的相关度达到0.82,能有效消除伪装效果评估过程中由操作人员主观因素对评估结果造成的影响,具有良好的科学性和可靠性。The eight-connected neighborhood,which takes the contour of the object as mask and centroid as origin,was defined as the immediacy backgrounds for target camouflage effectiveness evaluation.Based on target identification mechanism and correlative camouflage principle,19 characteristic values of the target and its 8 immediacy backgrounds were figured out from its statistic,shape and texture category respectively.Then these parameters were normalized by 3σ principle and an evaluation matrix of nine-rows and nineteen-columns was made out.By BP neural network,a quantification model of camouflage effectiveness evaluation was created,which was trained,validated and tested with a great deal project data from camouflage effectiveness tests.The correlation between this model and expert evaluation is 0.82,which indicates that the model is scientific and credible,and can effectively remove operators′ subjective factor in the camouflage effectiveness evaluation.

关 键 词：伪装评估 BP神经网络 特征提取 特征分析 

分 类 号：TP389.1[自动化与计算机技术—计算机系统结构]