Research On Adaptive Detection Method Of Ship Target In Remote Sensing Image

Ships are the main means of transportation at sea.The automatic detect ion of ship targets is of great significance in civil fields such as traffic command,maritime rescue and military fields such as reconnaissance and surveillance,military prediction and so on.With the advantages of high spatial resolution,large field of view and all-weather,optical remote sensing satellite has become an ideal way to obtain information of ship target detection task.At present,the main difficulties of ship target detection algorithm in optical remote sensing image include:(a)the land background of remote sensing image is complex,which seriously restricts the performance of ship target detection algorithm.(b)Ship target has the characteristics of multi-scale and multi direction,so it is necessary to find a more robust target detection method.(c)In different imaging scenes,the imaging angle,image resolution,signal-to-noise ratio and background type of remote sensing image are very different.To solve the problems,this paper focuses on the sea land segmentation method and ship target detection method :(1)Sea land segmentation based on global coastline data.Based on the time complexity and space complexity,this paper optimizes the data storage structure and designs the sea land segmentation method.Firstly,the global coastline database is established by using prior informa tion.Secondly,the related coastline in the database is mapped to the image coordinate system to complete the coastline matching.Finally,the land area is filled to complete the sea land segmentation,which avoids the influence of land area on ship targe t detection.(2)Research on the detection method of ship target.Firstly,the gray scale and shape of the ship target are analyzed.Considering the characteristics of multi-scale and multi-directional,a new method is proposed,which uses morphological filtering to obtain suspected targets,then calculates radial gradient histogram(RGT)features of suspected areas,and finally uses support vector machine to eliminate fal se alarms.In order to improve the adaptability of the target detection method to comp lex imaging scene and heterogeneous data detection task,an adaptive support vector machine model is proposed,which can adjust the position of hyperplane according to th e complexity of the image,and improves the performance and practicability of the detection method.(3)Experimental verification.Data sources are DOTA,hrsc2016 and Airbus,and accuracy,recall and F1 value are used as evaluation indexes.The results sho w that the average accuracy rate of the proposed algorithm is 81.67%.In order to verify the effectiveness of MDR adaptive support vector machine in complex scenes and heterogeneous data detection,ablation experim ents are set up,and the same data is used to train and test the detection methods of cascaded MDR adaptive support vector machi ne and traditional support vector machine.The results show that the proposed adaptive support vector machine has better detecti on effect.