Sea Ice Classification Of Remote Sensing Image Based On Neighborhood Relationships

Sea ice is one of the major factors affecting global climate and has different degrees of harm to high-latitude shipping,oil and gas exploration and production.Meanwhile,sea ice can generate various extent damage to the production and life in the coastal areas of the yellow sea and Bohai sea.In order to reduce the negative effects of sea ice,the accurate extraction of sea ice monitoring parameters such as sea ice type is necessary for both scientific research and production and life.Remote sensing technology can obtain long-term,continuous,large-scale and real-time monitoring images.Moreover,during the occurrence and development of sea ice,the situation and process of sea ice are monitored all day and all day,and then the sea ice situation can be obtained of a comprehensive,rapid and accurate assessment.Therefore,remote sensing images become the main means to obtain sea ice information of sea ice type.With the neighborhood relationship of sea ice,this paper studies and explores three methods of sea ice classification for remote sensing images.Firstly,we propose a novel sea-ice classification framework based on locality preserving fusion of multi-source images information.The locality preserving fusion arises from two-fold,i.e.the local characterization in both spatial and feature domains.We commence by simultaneously learning a projection matrix,which preserves spatial localities,and a similarity matrix,which encodes feature similarities.We map the pixels of multi-source images by the projection matrix to a set fusion vectors that preserve spatial localities of the image.On the other hand,by applying the Laplacian eigen-decomposition to the similarity matrix,we obtain another set of fusion vectors that preserve the feature local similarities.We concatenate the fusion vectors for both spatial and feature locality preservation and obtain the fusion image.Our locality preserving fusion framework is effective in classifying multi-source sea-ice images because it not only comprehensively captures the spatial neighboring relationships but also intrinsically characterizes the feature associations between different types of sea-ices.Secondly,we propose a novel sliding ensemble strategy,i.e.Sliding-Bagging,for sea ice classification.We commence by spatially dividing image plane into overlapping blocks of pixels.For each block,we use randomly selected pixels to train an individual classifier.Therefore,one pixel is covered by multiple blocks.In the classification procedure,one pixel is possibly assigned to multiple labels because each block covering the pixel assigns it a label based on he associated classifier.Following the bagging scheme,the final label of the pixel is the one that repeats with the most times.The proposed sliding ensemble strategy is effective for sea ice classification,because it not only avoids the limitation of one single classifier for characterizing the sea ice variability over a large-scale region,but also alleviates the data imbalance over different sea ice types.Finally,we propose a neighborhood relationships auto-context method for sea ice classification.In this method,we developed a novel method for context feature extraction,which used the context location point set to fit the pixels block region that the current pixel belongs.Then,the feature vector of spatial neighborhood relationship is extracted based on the position point set coordinates and the category.Moreover,following the auto-context method,the classifier is training by iterative method.The proposed neighborhood relationship autocontext is effective for sea ice classification,because the pixel intensity and texture feature are easily affected by external parameters,and spatial neighborhood relationship is more suitable for sea ice classification.In addition,we extract the spatial neighborhood relationship for each pixel,this method is more suitable for sea ice classification.Overall,neighborhood relationships is an important feature of sea ice,and effective use of neighborhood relationship can make sea ice remote sensing image classification have better classification accuracy.