基于偏振成像的水上航行场景语义分割方法  

Semantic Segmentation Method of Water Navigation Scene Based on Polarization Imaging

作　　者：肖长诗[1,2] 姜阳 文元桥 XIAO Changshi;JIANG Yang;WEN Yuanqiao(School of Navigation,Wuhan University of Technology,Wuhan 430063,China;Hubei Key Laboratory of Inland Shipping Technology,Wuhan 430063,China)

机构地区：[1]武汉理工大学航运学院,武汉430063 [2]内河航运技术湖北省重点实验室,武汉430063

出　　处：《武汉理工大学学报（交通科学与工程版）》2023年第5期956-962,共7页Journal of Wuhan University of Technology（Transportation Science & Engineering）

基　　金：国家自然科学基金(52001241)。

摘　　要：文中提出一种基于偏振成像的水上航行场景图像语义分割方法.利用偏振相机阵列图像采集设备,采集同一场景不同偏振方向的水上航行场景图像,合成偏振度图像,建立偏振图像数据集,构建基于偏振成像的图像语义分割网络模型.使用人工标注的偏振图像数据集开展训练,得到恶劣航行场景下的场景分割图像,并将结果与常用语义分割方法进行比较.结果表明:在恶劣航行场景下,能够获得较好的分割效果.A semantic segmentation method of water navigation scene image based on polarization imaging was proposed.Using the self-designed polarization camera array image acquisition equipment,the water navigation scene images of the same scene with different polarization directions were collected to synthesize polarization images.The polarized image data set was established,and the image semantic segmentation network model based on polarized imaging was constructed.The manually labeled polarized image data set was used for training,and the scene segmentation image under bad navigation scene was obtained,and the results were compared with the common semantic segmentation methods.The results show that it can achieve better segmentation results in bad navigation scenes.

关 键 词：水上航行场景 图像语义分割 深度学习 可见光偏振 

分 类 号：U664.82[交通运输工程—船舶及航道工程]