基于pix2pix的数码迷彩方案研究  

Research on Digital Camouflage Scheme Based on pix2pix

作　　者：冉建国 刘珩 张月 RAN Jian-guo;LIU Heng;ZHANG Yue(Army Engineering University of PLA, Nanjing 210007,China)

机构地区：[1]陆军工程大学,江苏南京210007

出　　处：《指挥控制与仿真》2022年第3期116-121,共6页Command Control & Simulation

摘　　要：由于光照、温度、天气和季节的不同,同一地域通常在外观上有很大的差异,颜色的多样性是设计数码迷彩伪装方案的主要挑战之一。针对迷彩伪装过程中目标与背景融合难的问题,开展了基于pix2pix网络进行数码迷彩研究,采用STST(stain-to-stain Translation)方法对背景图像进行颜色归一化,在保留相应的纹理特征基础上,通过对特定颜色分布在“pix2pix”框架下进行深度学习生成数码迷彩图案。在算法上和实验上相较现有的数码迷彩设计,本文生成的数码迷彩图案与背景融合度明显提高。Due to different light,temperature,weather and season,the appearance of the same region usually varies greatly.The diversity of colors is one of the main challenges in designing digital camouflage schemes.In order to solve the problem that it is difficult to fuse the target and background in the camouflage process,this paper carries out the research of digital camouflage based on pix2pix,and uses STST(stain to stain translation)method to normalize the color of the background image,which not only retains the corresponding texture features,but also distributes the specific color in"pix2pix"generate digital camouflage patterns through in-depth learning under the framework.Compared with the existing digital camouflage design in algorithm and experiment,the fusion degree of digital camouflage pattern and background generated in this paper is significantly improved.

关 键 词：数码迷彩 颜色归一化 Stain-to-Stain翻译(STST) pix2pix 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]