Research Of Complex Scene Ship Detection In Optical Remote Sensing Images

With the development of optical satellite imaging technology,the very high resolution and more target visualized features optical remote sensing images make ship detection become a hot research field.Now,in ship detection application,the state of the art method cannot meet practical system requirements under the complex scene situations.Therefore,this paper focuses on false alarms producing and target missing problems of ship detection under complex optical remote sensing scenes to carry out the research.The specific research work includes two aspects followed by the general remote sensing images processing procedure.One is scene level pre-processing,and another is target level interpretation.First,homogenous area extraction is employed to avoid false alarms producing and unnecessary redundant detection operations,which is related to optical remote sensing images widely covering and abundant land-use information containing characters.Furthermore,the homogenous area extraction can improve the timeliness of ship detection and lay the foundation for target level interpretation.The concrete scene level pre-processing includes the following two aspects:(1)Refined sea-land segmentation: Geographic information database(Note: DEM database 16 m resolution)spatial resolution cannot meet the refined sea-land segmentation demand caused by the improved very high resolution optical satellite images.Hence,this paper proposed a multi-scale pooling and features fusion image sequence generation for MRF(Markov Random Field)modeling method to achieve refined sea-land segmentation.The proposed method has realized preventing a huge of false alarms interference of ocean ship detection in the land area,and laid the foundation for harbor area extraction.(2)Harbor area extraction: Harbor is the important area where optical satellite are observed.Because the harbor area complex structure,which is difficult to use effective feature description to extraction the harbor area from a whole optical remote sensing scene.Therefore,this paper proposed an algorithm of pattern coding to achieve harbor area extraction,and the proposed algorithm realized the robust description and accurate extraction of the complex port area structure.Then,it lays a foundation of ship detection in harbor area.On the basis of scene pre-processing in optical remote sensing image ship detection,the target level interpretation includes two aspects:(1)Medium and low resolution optical remote sensing ocean ship detection: this paper focuses on the addressing the boundary problem(i.e.large area cloud covering,broken cloud,island interference,ship’s own gray and scale discrepancy change,algorithm real-time processing demand and so on.)of optical ocean ship detection under complex scenes.This paper proposed a hierarchical and differential modeling ship candidate extraction and comprehensive decision analysis candidate identification methods.It solves the problems of false alarms producing and ship target missing under complex ocean ship detection scenes,and proposed method has better robustness and adaptability.(2)High resolution optical remote sensing inshore ship detection: Due to the difference of inter-class,the multi-directional and multi-scale characteristics of high resolution inshore ships,and the complicated and varied interference information in the harbor background,which is the mainly reasons of false alarms producing and target missing problems in inshore ship detection application.This paper is aim at addressing the problems mentioned before and follows the demand of intelligent algorithm designing.The structural sparse representation for inshore ship region proposal method,common sharing sparse representation for ship’s orientation prediction and comprehensive structure saliency analysis for refined inshore ship’s contour detection methods are proposed for refined inshore ship detection.The proposed methods solve the existing problem of the state of the art methods for high resolution inshore ship detection,and realize the fine detection of inshore ships in high resolution optical remote sensing complex harbor scenes.This paper followed by scene level pre-processing to target level interpretation procedure,and studied ship detection in optical satellite images.The algorithm is given by solving existing problem in the sate of the art methods.A large number of experiments verified the efficitivency and adaptability of proposed algorithm,which lays the foundation for practical ship detection system application.