公里级背景数据采集及伪装特征提取方法  

Kilometer-level background data acquisition and camouflage feature extraction method

作　　者：杨鑫 许卫东 郝有斌 刘朝畅 YANG Xin;XU Weidong;HAO Youbin;LIU Chaochang(Field Engineering College,Army Engineering University of PLA,Nanjing 210007,China)

机构地区：[1]陆军工程大学野战工程学院,南京210007

出　　处：《兵器装备工程学报》2023年第1期144-151,共8页Journal of Ordnance Equipment Engineering

摘　　要：背景采集与处理是伪装设计的首要步骤。为了提高公里级背景范围采样的准确性和完整性,利用系统抽样理论提出了基于空中成像的背景两级采样方法。第一级采样通过剖分背景区域网格获取背景地物和地貌分布规律。第二级采样依据“接受-拒绝”法有针对地获取背景细节图像。基于颜色、距离和纹理特征差异性度量构建背景特征提取模型。然后通过SLIC(simple linear iterative clustering)算法分割超像素图像形成背景特征区域划分。实验中收集了某地300 km^(2)范围12个一级采样点的数据,并分析了背景地物占比和分布特性。针对分析的结果,二级采样采集了戈壁滩、农田村庄和道路三类背景数据。对比了不同分割参数下农田村庄背景图像的分割性能,并探讨了基于k-means和超像素分割模型的差异性。最后生成了3类背景的斑点形状和7种主色特征。结果表明,采样流程能够有效获取公里级背景数据分布和细节纹理颜色特征。超像素分割模型能够优化斑点区域的完整性。该研究为移动目标伪装设计提供基础依据。Background acquisition and processing is the first step of camouflage design.In order to improve the accuracy and integrity of kilometer background range sampling,this paper proposes a two-stage background sampling method based on aerial imaging by using the system sampling theory.The first level sampling obtains the distribution law of background features and landforms by dividing the background area grid.The second level sampling obtains background detail images according to the“Accept-Reject”method.A background feature extraction model is constructed based on the difference measurement of color,distance and texture features.Then,the super-pixel image is segmented by simple linear iterative clustering(SLIC)algorithm to form the background feature region division.In the experiment,the data of 12 primary sampling points in a range of 300 km^(2) are collected,and the proportion and distribution characteristics of background features are analyzed.According to the analysis results,the second level sampling collects three kinds of background data:Gobi desert,farmland villages,and roads.The segmentation performance of farmland village background images under different segmentation parameters is compared,and the differences between k-means and super-pixel segmentation models are discussed.Finally,three types of background spots and seven main color features are generated.The results show that the sampling process can effectively obtain the kilometer level background data distribution and detailed texture color features.The super-pixel segmentation model can optimize the integrity of spot regions.This study provides a basis for the camouflage design of moving targets.

关 键 词：背景采集 伪装特征 系统抽样 聚类分析 主色提取 

分 类 号：E951.4[军事—军事工程]