摘要
遥感卫星能够对云雾进行大范围、长时间监测,对海雾识别研究具有重要的意义,本文根据不同云、雾和下垫面的光谱特性及纹理特征,通过葵花8号卫星(Himawari-8)提取云图特征、CALIOP星载激光雷达(cloud-aerosol Lidar with orthogonal polarization,CALIOP)获取中高云、低云、海雾、晴空海表四类样本标签,利用深度学习方法构建深度神经网络(Deep Neural Networks,DNN),实现了白天海雾的有效识别。实验结果表明:本文训练得到的深度神经网络模型准确率为82.63%,具有较高的识别精度,而且相比其它海雾识别方法也有更好的识别结果。利用该模型对2016年4月8日发生于黄渤海区域的海雾天气进行识别,并通过CALIOP标签数据对识别效果进行验证,结果表明该方法能够较好的识别出海雾区域。
The remote sensing satellite can monitor cloud and fog in a large range and continuous time,which has great significance for the research of sea fog recognition.In this paper,according to spectral characteristics and texture features of different clouds,fog and underlying surface,relevant features are extracted from himawari-8 data,and four kinds of sample labels are obtained from CALIOP data,the samples include middle/high clouds,stratus,sea fog and clear sky sea surface.Then deep learning method was applied to construct deep neural network for achieving effective recognition of sea fog in daytime.The experimental results show that the accuracy of our model is 82.63%,which indicates it has a good performance,and the accuracy is also higher when compares with other sea fog recognition methods.The model is used to recognize sea fog in the Yellow Sea and the Bohai sea on April 8,2016.The results through verification using CALIOP data show that the method can recognize sea fog area well.It proves that the method proposed in this paper is feasible.
作者
司光
符冉迪
何彩芬
金炜
SI Guang;FU Ran-di;HE Cai-fen;JIN Wei(Faculty of Electrical Engineering and Computer Science,Ningbo University,Ningbo 315211,China;Zhenhai Meteorological Observatory,Ningbo 315202,China)
出处
《光电子．激光》
EI
CAS
CSCD
北大核心
2020年第10期1074-1082,共9页
Journal of Optoelectronics·Laser
基金
宁波市自然科学基金(2019A610104)资助项目。
