| With the development of the satellite remote sensing technology,it has been widely used to monitor the large area of the ocean environment.To effectively detect sea targets from satellite remote sensing images with large amount of data,it is important to distinguish the marine and non-oceanic regions(land,clouds,air flying objects,etc.)accurately.The main work of this paper is as follows.1.The statistical distribution characteristics of the sea and land in marine remote sensing images are analyzed.The features of local information entropy,local edge and local texture were introduced for the analysis of the statistical characteristics of the sea and land in remote sensing images.And it verifies the feasibility of ocean modeling in the three-dimensional feature space formed by the above features.This work also provides the basis for subsequent large scale ocean background modeling.Besides,the statistical probability density of the local land and sea is analyzed,and the characteristic distributions of the local regions are more similar if the regions are closer.So,it provides the basis for the extraction of the training samples.2,Aiming at the problem of sea-land segmentation for the large-field marine remote sensing image,the sea background modeling algorithm based on the multi-feature is proposed.Firstly,the sea-land is roughly segmented by using the global shoreline database GSHHS.On the basis of it,the paper presents a method for automatic extraction of coastlines based on low-frequency sampling technology.The inaccurate coastline margins extracted in the rough segmentation are as test samples for fine segmentation.Secondly,the paper proposes a automatic selection method of sea background based on FCM(Fuzzy C-means),to extract the pure seawater area for background modeling.Then,the multi-Gaussian seawater background modeling method based on three-dimensional feature space is presented.And the fine segmentation of the coastline edge region is achieved.Finally,the rough and fine segmentation results are fused.Experimental results show that the proposed algorithm has high segmentation accuracy and wide applicability,and is suitable for large area ocean remote sensing images with complex background.3,The statistical distribution characteristics of sea-cloud in marine remote sensing images were analyzed.First,the multi-attribute features of sea-cloud are introduced and the statistical probability density of the sea-clouds is analyzed.Then,this paper proposes a separability measurement method based on probability distribution curves,for the separability of sea-cloud features.Finally,for the feature description problem of sea-cloud,the first three features with higher separability are combined into a three-dimensional feature vector for statistical analysis.And the feasibility of sea background modeling in the 3D feature space is verified,which provides a basis for subsequent large-scale ocean background modeling.4.To solve the problem of sea-cloud segmentation of the large-filed ocean remote sensing image,a sea-cloud segmentation algorithm based on local sea background modeling is proposed.Firstly,this paper proposes a classification method for pure and impure seawater based on FCM,which is used to classify the subimages.Second,a seawater background modeling method based on multi-Gaussian in three-dimensional feature space is presented.Also,a sea sample extraction method based on the distance measure is given.Then,the sea background of the impure seawater subregions is modeled to segment the sea cloud.Finally,the segmentation results of the impure seawater subimages are merged.Experimental results show that the proposed algorithm can effectively distinguish the sea-cloud,and is suitable for various types of images.Finally,the work of this paper is systematically summarized,and prospects for improvement and further research are also given. |
