Research On Automatic Recognition Method Of Ship Type Based On Big Data Of Trajectory

With the mandatory installation of automatic identification system(AIS)equipment for ships,a large amount of ship AIS data is being generated every day.Ship type recognition based on ship AIS trajectory big data refers to the use of trajectory data in ship AIS data to identify the type of ship.Ship type identification can provide technical support for further research on camouflage ship identification,ship trajectory data pattern mining,and abnormal point detection.Based on the ship’s AIS trajectory data,this paper realizes the classification of five types of ships: fishing,cargo,container,passenger and tanker.The specific work is as follows:1.Through analysis,it is found that the ship’s AIS trajectory data has problems such as large data volume,data redundancy,disorder,and abnormality.In response to these problems,based on Hadoop and Spark and other big data components,the preprocessing method of ship’s AIS trajectory data is designed and implemented.It includes operations such as sorting the original ship’s AIS data,deleting abnormal data,data deduplication,sampling interpolation,calculating speed,heading,angular velocity,etc.,and finally converting it into trajectory image data.Then it is proved through experiments that this preprocessing method improves the performance of ship’s AIS trajectory data.Effectiveness provides a guarantee for the feature mining work;2.The previous feature extraction methods have insufficient in-depth problems,such as extracting features based on the entire trajectory,without considering that different ships may have more obvious differences in different sailing states,which leads to poor classification results.This article is based on Hadoop Big data components such as Spark and Spark divide the ship’s AIS trajectory data into three types: straight section,turning section,and stay section,and extract 63 statistical features such as speed,heading,and angular velocity from the three trajectory sections.3.The choice of model will affect the final classification results.This article trains ten machine learning models based on statistical features.The experimental results show that the KNN model and Decision Tree model have the best classification performance,and the F1 values of the two models have reached more than 90 percent.At the same time,a deep learning model,Res Net50 model,was trained based on the trajectory image data.The experimental results show that the F1 value of Res Net50 has reached more than 95%.Compared with the KNN model and the Decision Tree model,Res Net50 has a greater lead.4.Designed and implemented a ship type automatic identification system based on ship AIS trajectory big data.Based on the B/S architecture,the system is divided into four functional modules: data preprocessing,statistical feature mining,model training,and ship type recognition.