Research On Techniques Of Ship Detection On The Sea Surface In Remote Sensing Images

Marine ship detection has always been a hot issue in the field of remote sensing,and related detection technologies are widely used in the field of national maritime security and marine economics.In terms of maritime safety,ship detection can be used to monitor the position and movement of enemy ships and provide information for subsequent military decision-making;in terms of marine economy,the dredging of shipping,the management of ocean fishing vessels,and the rescue of ships in distress,etc.,ship detection Technology can also play a huge role.However,the large acquisition area of remote sensing images and the complex background environment of the images have caused many difficulties to ship detection,such as the interference of clouds,land,and the existence of reflections from the sea surface,and other man-made objects.Therefore,this paper designs an accurate and fast ship Detection system based on region of interest detection and marine targets classification in remote sensing images.First,in order to improve the quality of remote sensing images and improve the overall efficiency of the Detection algorithm,the input remote sensing images are preprocessed.Improve image quality by stretching transformation,median filtering and down sampling,an adaptive threshold mask based on Otsu has been used to eliminate irrelevant areas such as land and clouds,and uses mathematical morphology to refine the mask boundary.Experiments prove that the series of preprocessing in this paper improve the quality of remote sensing images and the accuracy of subsequent Detection.Then,in order to improve the accuracy of ship detection,region of interest detection for target is proposed.Based on the classical Gauss-Laplace operator,it is proposed to detect bright spots in the image with the negative function of the Gauss-Laplace,and combined with the actual difference of spot size,obtain the response of different spots with a multi-scale detection method.But the spot response alone is difficult to locate the target position accurately.Therefore,this paper proposes a region-of-interest localization algorithm based on strongly connected components to analyze the response information and extract the accurate coordinates and scale information of the target.Experiments prove that the improved Gauss-Laplace bold detection combined with the region-of-interest localization algorithm based on strongly connected components can accurately locate the locations of region of interest,and crop samples of targets in remote sensing images.Finally,on the basis of accurately locating the region of interest,the cropped target samples include not only ships but also other maritime objects such as breeding cages.In order to distinguish these samples and realize the detection of ships,this paper designs the classification of remote sensing maritime targets.Using an improved VGGNet to reduce the parameters of the classification network and improve the operating efficiency.In order to further improve the accuracy of the network classification,based on the 8-layer VGGNet,this paper uses the residual network as the target sample classification network,and uses the commonly used Res Net18 and Res Net34 as alternatives,and uses the artificial remote sensing image data set to design comparison experiments.According to the test accuracy of VGGNet8,Res Net18 and Res Net34,the 34-layer residual network structure is selected to design the classification algorithm.A large number of experiments have proved that the network structure can well complete the classification task of marine targets,distinguish ships from other marine targets,and finally realize the detection of remote sensing image marine ships.