基于GAN的船舶遥感图像数据增广方法  

Data augmentation method of ship remote sensing images based on GAN

作　　者：杨志钢 杨远兰 苍思远 李泳江 郝燕云 张帆[3] 吴根水[3,4] YANG Zhigang;YANG Yuanlan;CANG Siyuan;LI Yongjiang;HAO Yanyun;ZHANG Fan;WU Genshui(Key Laboratory of Advanced Marine Communication and Information Technology,Ministry of Industry and Information Technology,Harbin Engineering University,Harbin 150001,China;College of Information and Communication Engineering,Harbin Engineering University,Harbin 150001,China;China Airborne Missile Academy,Luoyang 471009,China;Aviation Key Laboratory of Science and Technology on Airborne Guided Weapons,Luoyang 471009,China)

机构地区：[1]哈尔滨工程大学先进船舶通信与信息技术工业和信息化部重点实验室,黑龙江哈尔滨150001 [2]哈尔滨工程大学信息与通信工程学院,黑龙江哈尔滨150001 [3]中国空空导弹研究院,河南洛阳471009 [4]航空制导武器航空科技重点实验室,河南洛阳471009

出　　处：《应用科技》2022年第5期8-14,23,共8页Applied Science and Technology

基　　金：航空科学基金项目(201801P6002)。

摘　　要：为了解决光学遥感探测中民用商用船舶和军舰样本不足的问题,提出面向船舶遥感图像的Wasserstein距离生成对抗网络(ship-WGAN)。在ship-WGAN中,生成器网络以U-Net结构为主,并利用重建损失和感知风格损失提高生成器生成图像的能力;在判别器中引入残差模块,减少网络参数的计算量。在Google Earth获取5类船舶遥感图像建立训练集,以此验证ship-WGAN的性能;在生成图像质量方面,峰值信噪比、结构相似性度量、起始距离分别可以达到24.91 dB、0.92、0.99;扩充样本后训练图像识别模型,识别准确率能够提升6%。提出的ship-WGAN可以生成高质量的船舶遥感图像虚拟样本,提高船舶识别的准确率,达到数据增广的效果。In order to solve the problem of insufficient samples of civil and commercial ships and warships in optical remote sensing detection,this paper proposes a Wasserstein distance generative adversarial network(ship-WGAN)for ship remote sensing images.In the proposed ship-WGAN,the U-Net structure is used as the generator network,and the reconstruction loss and perceptual style loss are introduced to improve the ability of generators to generate images.Furthermore,the residual module is introduced into the discriminator to reduce the amount of calculation of network parameters.And a training set is established by obtaining five types of ship remote sensing images in Google Earth to demonstrate the effectiveness of the proposed method.The peak signal to noise ratio,structural similarity and frec het inception distance of the generated images can reach 24.91dB,0.92 and 0.99,respectively.The recognition accuracy can be increased by 6%after increasing samples and training the image recognition model.The proposed ship-WGAN can generate high-quality and realistic ship remote sensing images,improve the recognition accuracy,and achieve the effect of data augmentation.

关 键 词：船舶遥感图像 数据增强 生成模型 生成对抗网络 图像生成 Wasserstein距离 图像分类 计算机视觉 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]