Remote Sensing Ship Detection Based On Significant Candidate Regions

In recent years,With the continuous breakthrough of remote sensing technology,Thanks to a large number of high-resolution remote sensing satellites,optical remote sensing images can obtain more abundant information,and ship detection has more diversified remote sensing image data.With the increasing application requirements,optical remote sensing ship detection methods with more efficient and higher recognition capability have important research significance and application value.For this reason,the research topic of remote sensing ship detection algorithm is carried out,which includes two aspects that one is the extraction of significant candidate regions and another one is the identification of ship candidate regions.Firstly,the characteristics of optical remote sensing images and the principle of saliency detection model are introduced.Meanwhile,fitting experiments and saliency detection experiments on different characteristics of optical remote sensing images and several common saliency models are carried out respectively,which lays a foundation for subsequent optical remote sensing ship detection.Secondly,in order to solve the problem of low saliency value of ships similar to sea surface color in the optical remote sensing ship detection task scene,a saliency candidate region extraction algorithm that combines the improved FT saliency detection and Hessian edge detection saliency model is proposed.The improved FT saliency detection and Hessian matrix edge detection were used to process the remote sensing images separately first and then the pulse coupled neural network is used to fuse the obtained two saliency maps into a total saliency map,so as to improve the saliency value of ships with similar background colors to the sea surface,thus extracting effective ship candidate region slices,and normalizing the size of the candidate region slices,which is convenient for the next detection process.Experiments show that the algorithm effectively improves the saliency value of ships with similar background colors to the sea surface in the ship detection task.Finally,aiming at the problem of background interference in ship candidate region extraction,a candidate region identification method combined with migration VGG16 is proposed.The features of ship data set are extracted by migrating VGG16 network first and then Softmax classifier is trained to identify the candidate region to separate the possible interference information in the candidate region,thus realizing remote sensing ship detection.Experiments show that the algorithm further improves the accuracy of optical remote sensing ship detection.