基于微光图像和深度学习的海洋内波检测方法  

Deep learning-based detection of oceanic internal wave for satellite low-light image

作　　者：朱起文 胡申森 艾未华 郭朝刚 李浩 ZHU Qiwen;HU Shensen;AI Weihua;GUO Chaogang;LI Hao(University of Chinese Academy of Sciences,Shijingshan Beijing 100010,China;National University of Defense Technology,Kaifu Changsha,410073,China)

机构地区：[1]中国科学院大学,北京石景山100010 [2]国防科技大学,长沙开福410073

出　　处：《自动化与仪器仪表》2023年第9期16-20,共5页Automation & Instrumentation

摘　　要：海洋内波是一类在大陆架、海峡和岛屿附近海域常见的波动,可能对水下舰艇和海上石油平台等造成严重安全威胁。卫星微光图像是近年来用于检测夜间海洋内波的一种重要手段,而目前主要依赖人工判别的方式,利用人工智能技术从微光图像中自动判别海洋内波的研究还较少。针对海洋内波在微光图像中呈现明暗相间条纹这一特征,发展了一种适用于微光图像的海洋内波智能检测算法。利用2018年至2019年期间南海区域的VIIRS微光图像建立训练与验证数据集,并结合迁移学习和数据增强技术训练卷积神经网络Inception Net V3,得到基于深度学习的海洋内波检测模型。利用2020年的VIIRS数据进行检验,结果表明海洋内波检测准确率为81.34%。Ocean internal waves are common in continental shelves,straits,and islands,which may pose a serious security threat to underwater ships and offshore oil platforms.Satellite low-light image is an important means to detect the ocean internal waves at night,but at present,it mainly depends on the way of artificial discrimination.There are few studies on automatic discrimination of ocean internal waves from low-light images using artificial intelligence technology.Because the ocean internal wave presents light and dark stripes in low-light images,this paper developed an intelligent detection algorithm of ocean internal wave for low-light images.The proposed algorithm used the VIIRS low-light images over the South China Sea region from 2018 to 2019 to establish a training and validation data set,and combines migration learning and data enhancement techniques to train the convolutional neural network Inception Net V3 to obtain the ocean internal wave detection model based on depth learning.The VIIRS data in 2020 were used for independent validation,and the results showed that the accuracy of ocean internal wave detection is 81.34%.

关 键 词：海洋内波 微光遥感 深度学习 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]