Classification Of Ship Navigation Behavior In Cross-channel Based On Sparse Representation

With the development of domestic water transport in our country,inland waterways are gradually congested,especially the frequent accidents of inland waterways,which bring great challenges to inland waterway management and waterway supervision.In recent years,with the application and popularization of automatic identification system(AIS)and data mining technology,the use of massive AIS data for ship behavior research and trajectory classification has become one of the hot research directions.This thesis focuses on the classification of ship navigation behavior,and identifies the steering behavior pattern of ships at the crossing channel by classifying the trajectory before the ship enters the cross-channel,which plays a certain role in auxiliary decision-making in the fields of traffic flow statistics,auxiliary maritime supervision and water traffic planning.However,in the application process,there are still some problems in AIS data that need to be solved,including many abnormal values and serious data loss.In view of the existing foundation and existing limitations of the current ship navigation behavior classification and trajectory anomaly screening,this thesis applies the sparse representation classification method to the field of ship behavior classification,and proposes a ship navigation behavior classification model based on sparse representation and dictionary learning.Firstly,a method combining fixed threshold and adaptive threshold is proposed to screen trajectory outliers.This method set a fixed threshold based on channel geographic information to screen out longitude and latitude anomalies in AIS original data.The trajectory of the sliding window is set at different trajectory,and the threshold of the sliding window is set adaptively according to the speed and course in the sliding window.After processing the anomalies,three data sets of cross channel,Y channel and T channel were made for experimental verification.Then,aiming at the problem of "same features and different types" of ship trajectories,this thesis uses cd SRC(Class-Dependent Sparse Representation Classifier)as the model.By combining traditional Sparse Representation Classifier(SRC)with improved K-Nearest Neighbor(KNN)model based on Mahalanthus distance,and proposed Mcd SRC(Mahalanobis Distance-class dependent Sparse Representation Classifier)algorithm,so that distance information between sample classes is introduced to improve classification accuracy.The improved algorithm uses the traditional SRC model to reconstruct samples,and updates the dictionary by K-singular value decomposition(K-SVD)algorithm to get the residual difference between the reconstructed samples and the samples to be classified,and then calculates Mahalanobis distance information by combining the improved KNN algorithm.The two are combined to form a classification discriminant function,which is classified by the principle of minimum error.Then,in order to solve the problem of large similarity between trajectory classes,the least squares cubic spline fitting algorithm is used to construct the characteristic matrix of ship navigation behavior based on trajectory points on the basis of the discriminant dictionary model,and Fisher criterion is added on the basis of the traditional sparse representation classification model,so that the divergence within different samples becomes smaller and the divergence between classes becomes larger.So as to improve the accuracy of classification.Finally,after processing anomaly points,three data sets of cross channel,Y channel and T channel were tested to verify that the above two classification frameworks can effectively improve classification efficiency.This thesis uses two improved models based on sparse representation classification model as classification algorithms to identify different behavior patterns of ships by classifying the trajectory of ships in the specific scenario of crossing channels,so as to have certain auxiliary value in maritime supervision,maritime traffic control,risk assessment,abnormal trajectory screening or traffic statistics.