基于深度卷积对抗网络的迷彩设计方法  

Camouflage Design Method Based on DCGAN

作　　者：冉建国 刘珩 张品 刘亚文 吕振坚 RAN Jianguo;LIU Heng;ZHANG Pin;LIU Yawen;LYU Zhenjian(National Key Laboratory on Electromagnetic Environmental Effects and Electro-optical Engineering, Army Engineering University of PLA, Nanjing 210007, China)

机构地区：[1]陆军工程大学电磁环境效应与光电工程国家级重点实验室,南京210007

出　　处：《兵器装备工程学报》2021年第11期264-269,共6页Journal of Ordnance Equipment Engineering

摘　　要：利用深度卷积生成对抗网络,构建特征空间,改进目标函数,优化算法,建立迷彩设计伪装中目标与背景图像之间的特征相似度的数学模型,对不同背景的图像特征数据进行自主对抗训练,形成与该背景特征高度匹配的图像,从而达到良好的伪装设计效果。结果表明:采用五维特征空间为目标函数构建的优化算法,对林地背景图像训练后的SSIM值90%以上有所提高,生成图像中的目标与背景融合度高,伪装效果好。In camouflage design,the target and background are difficult to be highly integrated.Based on the deep convolutional generative adversarial networks(DCGAN)to fix the problem,we constructed the feature space to improve the objective function of the network,and established the mathematical model of feature similarity between target and background;we got the training image feature data of different backgrounds independently and achieved good camouflage design effect.A similar image highly matched with the background feature was formed.The results show that the SSIM value of woodland background image was improved by more than 90%,and the target has a high fusion with the background,achieving excellent camouflage effect.

关 键 词：生成对抗网络 高融合迷彩设计 卷积神经网络 JS散度 特征空间 

分 类 号：TP394.1[自动化与计算机技术—计算机应用技术]