摘要
海雾无论在海上还是在沿岸地带,都因其恶劣的能见度对交通运输、海洋捕捞和海洋开发工程以及军事活动等造成不良影响,因此对于海雾的实时监测和预报就显得尤为重要。本文提出了基于深度学习的静止气象卫星多通道图像融合分割算法,使用D-LinkNet深度卷积神经网络语义分割算法模型对黄渤海海域范围的16通道、空间分辨率为0.5 km的Himawari-8卫星数据进行研究。分别采用均交并比(m_(IOU))以及观测值检验作为评价指标,在测试集上的m_(IOU)为0.9436,并且用卫星测试数据结果与海上观测数据结果进行对比,得出雾区准确率(检测有雾且真实有雾/检测有雾)为66.5%,雾区识别率(检测有雾且真实有雾/(真实有雾-云覆盖))为51.9%,检测正确率(检测正确/总样本)93.2%。本文提出的方法能为海雾监测提供一个可靠的参考。
Sea fog, whether on the sea or the coast, has adverse effects on transportation, marine fishing, marine development projects, and military activities due to its poor visibility. Therefore, real-time monitoring and forecasting of sea fog are essential. This paper proposes a multi-channel image fusion segmentation algorithm for stationary meteorological satellites based on deep learning. The D-LinkNet deep neural network semantic segmentation algorithm model is used to study the 16-channel Himawari-8 satellite data with a spatial resolution of 0.5 km in the Yellow Sea and the Bohai Sea. Using mIOU(mean Intersection Over Union) and observation value test as evaluation indicators, the mIOU on the test set is 0.9436, and comparing the results of satellite test data with the results of marine observation data. It was concluded that the accuracy rate of fog area(detect fog and real fog/detect fog) is 66.5%, the recognition rate of fog area(detect fog and real fog/(real fog-cloud coverage)) is 51.9%, and the detection accuracy rate(detection correct samples/total samples) is 93.2%. In conclusion, the method proposed in this paper can provide a reliable reference for sea fog monitoring.
作者
黄彬
吴铭
孙舒悦
赵伟
崔战北
吕成
HUANG Bin;WU Ming;SUN Shuyue;ZHAOWei;CUI Zhanbei;LYU Cheng(National Meteorological Center,Beijing 100081;Beijing University of Posts and Telecommunications,Beijing 100876)
出处
《气象科技》
2021年第6期823-829,850,共8页
Meteorological Science and Technology
