Ship Detection Based On Optical Remote Sensing Image

As an important research topic combining computer vision and remote sensing monitoring,ship detection technology has a wide range of application scenarios,including national territorial sea monitoring,maritime traffic management,and ship rescue.With the continuous progress of optical remote sensing technology in recent years,ship detection algorithms based on optical remote sensing images have received more and more attention.The detection accuracy and computational efficiency are key issues.This thesis analyses and studies the following aspects.Optical remote sensing images have problems such as target scale diversity,small target ratio,and arbitrary directionality.Traditional target detection algorithms are not suitable for ship detection scenarios.Considering the optical remote sensing image information,ship geometry and the characteristics of ship detection task,this thesis proposes a rotation area locating network and designs a feature pyramid layer to enhance the features fusion.The rotation anchor box strategy improves the quality of region proposals.Model optimization is performed on the aerial images data set.Experiments show that the comprehensive performance of the proposed model is superior to traditional algorithms.Especially in the complex scenes where ships are densely arranged,the proposed model predicts ship target position more precisely.Considering that redundant land background causes huge interference to the ship detection task and reduces the efficiency and accuracy of the algorithm,this thesis proposes a multi-level sea-land segmentation.The proposed algorithm uses digital elevation model to roughly distinguish between land and sea areas and uses fuzzy clustering to distinguish between marine area pixels and land area pixels.It uses morphological processing and isolated area erasure measures to ensure the accuracy of the final result.Experiments show that the proposed algorithm improves the miss detection rate,false detection rate,detection coverage rate,and detection accuracy rate,and has good robustness.