Application Of Deep Learning Method In Sea Fog Segmentation Scene Of Satellite Image

Sea fog is a kind of catastrophic weather that occurs in the sea and its coastal areas.It will cause great harm to the production and life on the sea.Since the early 1970s,scholars have devoted themselves to the study of sea fog characteristics and tried to find ways to accurately detect sea fog.The mainstream thinking in the field of sea fog detection is to conduct research based on the difference in cloud and fog radiation characteristics.The traditional methods proposed by scholars are mostly threshold methods that rely on experience.These methods often require a high level of expertise and fail to achieve satisfactory prediction accuracy.Deep learning methods have achieved great success in the field of computer vision.However,the application of deep learning techniques to sea fog detection is rare.Therefore,this article decided to use deep learning technology to detect sea fog in the Yellow Sea and Bohai Sea in China.The main content of this paper is divided into three parts:building a sea fog data set,training a deep convolutional network for sea fog prediction,and realizing an automated sea fog prediction system.In terms of data set construction,this paper established a highprecision and high-resolution sea fog satellite image data set.In this paper,the satellite remote sensing data of the Japanese Sunflower No.8 meteorological satellite is used as the data source,and the designated area is the Yellow Sea and Bohai Sea of China.We proposed two methods to produce high-precision label data,and based on these,we successfully constructed a sea fog data set with high labeling accuracy.In terms of network structure design,this paper tried the classic convolutional neural network in the segmentation field and selected DLinkNet as the final basic model.After that,this article optimized the network model,including the use of ASPP hollow space pooling pyramid to replace D-Block;the introduction of an attention mechanism,using the PAM module proposed by DA-Net,which significantly expanded the receptive field,Can capture sea fog information more comprehensively.Finally,this article has completely realized a set of automated sea fog detection system,which satisfies the needs of meteorological workers well.The system has been put into use and has been operating stably so far.The sea fog segmentation algorithm proposed in this paper achieves high accuracy and real-time performance on the self-built data set,which satisfies the meteorological staff’s demand for sea fog forecast to a high degree,and is also for the follow-up related sea fog field.The research work provides some ideas that can be used for reference.