The Research Of Remote Sensing Image Ship Detection Method Based On Visual Saliency

In recent years,with the continuous progress of science and technology in China,more and more remote sensing satellites have entered the space,which provides a huge amount of data support for researchers who apply remote sensing images to do scientific research,which also makes the research direction of remote sensing image target detection become more and more popular.Ship target detection has a wide range of application scenarios and has unique research value and significance.Due to the different shooting angle and time of the remote sensing image,as well as the situation of the ship’s sea area,the remote sensing image with ship target will have image color distortion,low color contrast between ship and sea surface,serious sea clutter or cloud occlusion,island and sea shore interference in some images,which makes ship detection process easy to miss detection or false detection.Most of the current ship detection methods,some detection results accuracy is not high,and some detection efficiency is low,which is very limited in practical application.Therefore,in this thesis,a ship detection method based on visual saliency is designed,which can remove false alarm and improve the efficiency and accuracy of ship detection.The specific research work of this thesis is as follows:1)Firstly,the mainstream methods of ship detection in remote sensing images are studied,and the overall framework of ship detection is analyzed and proposed.The ship detection is divided into two parts: rough detection and fine detection.Then,aiming at the feature that visual saliency model can effectively extract saliency information from the current image,four typical visual saliency models are obtained through experiments,and the advantages and disadvantages of each model are analyzed in depth.On this basis,a visual saliency model with frequency domain transformation is proposed.This model can effectively remove the negative effects of sea clutter,thin clouds and color distortion of remote sensing image when applied in marine remote sensing image,and we use this model to complete the rough detection stage of ship survey.2)After extracting the suspected ship area from remote sensing image by saliency detection,there are still some false alarms such as coastline,island,or heavy clouds in the slice of suspected ship area,although it can effectively suppress the interference of some image background information.In order to solve this problem,this thesis first uses clustering method to segment the slice,and then gets the object in the slice by removing the interference of a small amount of ocean background.Then,in order to solve the problem that the direction of the slice target is not unified,the slice target is placed in the vertical direction by using Hough transform.Finally,through the improved directional gradient feature and the introduction of identification parameters,the influence of false targets such as thick clouds and small islands is further eliminated to improve the accuracy of ship detection.The experimental results show that the ship detection method proposed in this thesis has a good effect in the visible remote sensing image with a resolution of 0.5 meters to 2 meters.In the stage of ship rough detection,the accuracy rate of detection results is 81.17%,and the recall rate is 95.29%.After the ship precise detection process,the accuracy rate of the final detection results of the ship detection scheme in this thesis is 90.04%,and the recall rate is 91.67%.