Study And Application Of Sea Fog Detection Model Based On Time Series And Deep Learning

Sea fog is a common meteorological phenomenon in Marine and coastal areas,which will have a great impact on maritime traffic,port transportation,coastal engineering and other fields.Therefore,the accurate detection and prediction of sea fog is of great significance for ensuring maritime safety and promoting economic development.The traditional sea fog detection method mainly uses meteorological observation data to predict,but this method has problems such as low accuracy and time delay.With the continuous development of computer technology,using deep learning models to achieve sea fog detection has become a new research direction.Combined with remote sensing satellite data,this paper carries out the research on daytime sea fog detection based on deep learning,and proposes a daytime sea fog detection model based on time series and improved U-Net.The main work of this paper is as follows:(1)In view of the lack of data sets for sea fog detection at present,this paper selected the Sunflower 8 satellite data as the original data,and took the Yellow Sea and Bohai Sea as the research area,carried out initial data collection,preprocessing and manual labeling,and constructed the original data set.The original data set is far from enough for deep learning model training,and this paper carries out the research of data augmentation method based on DCGAN.Finally,the sea fog data set required for the subsequent training of deep learning sea fog detection model is constructed.(2)Aiming at the problem of single-scale feature extraction and uniform image features in sea fog image segmentation,this paper proposes an improved U-Net network named DAU-Net.Firstly,asymmetric multi-scale convolution was proposed and introduced into U-Net network to realize multi-scale feature extraction.Secondly,the attention module is introduced into the U-Net network to highlight the significant features useful for sea fog segmentation,so as to improve the segmentation accuracy of the image segmentation model based on a single image.(3)Considering the evolution characteristics of sea fog,in order to make full use of the time series information of sea fog images,this paper combines Conv LSTM with DAUS-Net model,and performs sea fog detection by inputing sea fog image sequence.This method can make the sea fog detection make full use of the sea fog images before and after frames,and extract the temporal features in the image sequence,so as to detect the sea fog target more accurately.On the real sea fog dataset produced in this paper,the proposed model is compared with several models to verify the effectiveness of the proposed model.(4)In order to facilitate the sea fog detection for users,a sea fog detection system based on the sea fog detection model is developed.The analysis and design of the sea fog detection system follow the method of software engineering,and the implementation effect of each module is shown.Finally,the system function is tested to ensure its reliability and stability.