首页 >  , Vol. , Issue () : -

摘要

全文摘要次数: 629 全文下载次数: 858
引用本文:

DOI:

10.11834/jrs.20221621

收稿日期:

2021-09-23

修改日期:

2022-04-22

PDF Free   EndNote   BibTeX
注意力机制下多尺度特征融合生成对抗网络的白天海雾监测方法
方旭源1, 金炜1, 符冉迪1, 李纲1, 何彩芬2, 尹曹谦1
1.宁波大学信息科学与工程学院;2.宁波市镇海气象局
摘要:

海雾是海上一种常见的天气现象,它使能见度降低,给海上交通和作业带来极大威胁。传统的卫星遥感海雾监测算法在准确率、可移植性及自动化程度等方面都有待改善。本文在注意力机制下,利用卫星遥感云图,提出一种多尺度特征融合生成对抗网络的白天海雾监测方法。该方法引入通道注意力机制,通过学习不同输入通道的权重,提升了网络对于重要通道云图的关注度;在此基础上,采用多尺度特征融合以获取海雾的多尺度信息,使提取的特征能兼顾海雾的整体及细节特性;为了进一步提高算法对于对海雾边缘的界定能力,本文引入对抗网络对海雾监测的生成网络进行监督,从而得到更精细的海雾区域。在测试云图的海雾监测实验中,命中率(POD)、临界成功指数(CSI)及误报率(FAR)分别为90.5%、81.28%和10.86%,均优于传统海雾监测方法以及其它基于深度学习的方法,这表明本文方法可以有效提升海雾监测的精度,研究成果对于海上船只航行、渔业生产、国防军事等具有重要意义。

Daytime sea fog detection based on multi-scale feature fusion of generate adversarial network under attention mechanism
Abstract:

Sea fog is a common weather phenomenon at sea. It will reduce visibility at sea and pose a significant threat to maritime traffic and other operations. Traditional sea fog detection algorithms using satellite remote sensing have low accuracy, poor portability, and low automation. Although some existing deep learning-based sea fog monitoring algorithms have been improved, they do not consider the spectral characteristics of sea fog in different channels, and the accuracy of sea fog monitoring is not high, especially in edge recognition. In order to improve the accuracy of sea fog detection, a daytime sea fog detection method was proposed, which based on multi-scale feature fusion of generate adversarial network (GAN) under attention mechanism. Firstly, according to the spectral response of sea fog in different imaging channels of meteorological satellite, the satellite cloud images of different imaging channels that can reflect the characteristics of sea fog are selected as the input of the network. Meanwhile, in order to make the network better focus on the significant imaging channels under multi-channel input, the channel attention mechanism is introduced to measure the weights of different input channels. Then, in order to solve the problem of the loss detail features of the cloud image caused by the pooling operation of the traditional deep network, a multi-scale feature fusion mechanism is adopted to fuse the feature maps of different levels of the network to obtain the multi-scale features of the sea fog. Finally, in view of the difficulty of traditional methods to accurately describe the edge of sea fog, the generation network for sea fog detection is supervised by adversarial network, so as to accurately define the edge of sea fog and reduce the false alarm rate. This article takes the Yellow Sea and the Bohai Sea as the research area. Since March to June each year is the period of high incidence of sea fog in the Yellow Sea and the Bohai Sea, we produce a data set based on the weather satellite monitoring report of the National Meteorological Center from March to June 2017-2020. After the model training in this article, in terms of quantitative indicators of sea fog detection, probability of detection (POD), critical success index (CSI) and false positive rate (FAR) of our method are 90.5%, 81.28% and 10.86% respectively, which are better than other methods. The experimental results show that the method in this paper can effectively improve the accuracy of sea fog identification, which is of great significance for marine vessel navigation, fishery production, national defense and military affairs.

本文暂时没有被引用!

欢迎关注学报微信

遥感学报交流群