Research Of Inshore Ship Detection Based On High Resolution Optical Remote Sensing Imagery

Ship detection using remote sensing images has gained great attractions due to its application in maritime security,harbor traffic control and fishery management.But at present,most of the popular techniques for inshore ship detection are faced with several issues:(1)the wide scene in remote sensing images is very complex with multiple objects and messy background(2)in high-resolution images,the dock and other facilities in the harbor area are extremely similar to ships in gray scale and shape(3)there are multiple ship objects with different shape,direction and scale.To address issues above,we proposed a sparse representation based inshore ship detection technique.Our main contributions are in three-folds:(1)As most of the popular sea-land segmentation methods preform badly in the high-resolution remote sensing images,we proposed a pyramid integral image reconstruction algorithm for sea-land segmentation.We use a sum area table to compute the integral image,so as to extract the structural information from the image.By using multi-scale integral image analysis,we achieve good performance on sea-land segmentation problem.(2)We proposed a novel region proposal method based on sparse representation.First,we use PCANet to extract hierarchical features from the image.Then we design a multi scale and multi direction embedded dictionary to detect ships of different scale and angle.(3)We use sparse representation based classification method to further recognize ship object from the proposal candidates.By using label consistent K-SVD method,we are able to learn an over-complete dictionary with both reconstructive and discriminate ability,and it can effectively improve the classification accuracy.