Research On Data Mining Of Fishing Vessel Trajectory Based On Deep Learning

China is a big fishing country in the ocean.In recent years,with the increasing number of fishing boats,it has brought great challenges to fishery supervision departments,and at the same time brought serious damage to the marine ecological environment.With the rapid popularization of modern mobile intelligent terminals and the development of location sensing technology,it is easier and easier to obtain trajectory data including time,position,speed,acceleration and other characteristics.After the automatic identification system(AIS)based on satellite positioning technology has been widely deployed in the world,China has also deployed it accordingly and basically completed the terminal promotion of fishing vessels.AIS trajectory data,which contains dynamic information such as the position,speed and heading of fishing vessels,has exploded,which makes it possible to use AIS trajectory data of fishing vessels to monitor fishing vessels in real time,analyze the temporal and spatial distribution of fishery resources and protect the marine ecological environment.Therefore,based on the AIS trajectory data of fishing boats,this paper studies the mining from three aspects: AIS trajectory segmentation,fishing boat operation status identification and fishing boat operation mode identification.It is of great significance to store and calculate AIS trajectory data of fishing boats,standardize fishing behavior of fishing boats,ensure safety at sea and master the dynamics of fishery resources.The main work of this paper is as follows:(1)In the aspect of AIS trajectory segmentation of fishing boats,the segmented sub-trajectories contain different motion states,which is conducive to data storage,identification of fishing boat operation states and other data analysis.Aiming at the problems of existing trajectory segmentation algorithms,such as requiring prior values,labeling data and high computational complexity,this paper proposes an unsupervised trajectory segmentation algorithm based on multi-motion features(TS-MF),which includes two steps: segmentation and merging.In the segmentation,Pearson coefficient is used to measure the similarity of the features of adjacent track points,and the segmentation points are extracted from the global perspective.In the merging part,the minimum description length(MDL)value is optimized by merging local sub-tracks,which can avoid over-segmentation and improve the accuracy of track segmentation.The time complexity of TS-MF is O(n),which is suitable for large data sets such as AIS trajectory data of fishing boats.Finally,TS-MF is used to segment AIS trajectory data of fishing boats,and compared with four trajectory segmentation algorithms: GRASPUTS,SWS,CB-SMOT and SPD.The experimental results show that the harmonic average of TS-MF’s purity and coverage is the highest,which proves the effectiveness and reliability of the trajectory segmentation algorithm proposed in this paper.(2)Most of the existing algorithms are based on clustering,machine learning and empirical threshold,while few algorithms based on deep learning ignore the unequal lengths of sub-trajectory sequences.In order to solve this problem,this paper proposes an algorithm for identifying the fishing boat’s operation state based on long-short memory network(LSTM)and one-dimensional convolutional neural network(1DCNN).In this algorithm,the improved LSTM network is used to obtain the real time domain characteristics of the trajectory,and the real time step of each sub-trajectory is input at the input layer of improve LSTM network,and the time domain characteristics are output at the output layer of improve LSTM network according to the time step.Then1 DCNN network uses convolution and pooling operations to obtain the spatial characteristics of the trajectory.Finally,the output values of LSTM and 1DCNN are fused to calculate the value of the cross entropy loss function,which is continuously optimized until the result is optimal.The experimental results show that the accuracy rate of the proposed fishing boat operation status recognition algorithm reaches 96.2%,which is superior to other algorithms.(3)In the aspect of fishing boat operation mode identification,in order to extract the space-time characteristics of the trajectory and adapt to the processing of long-term trajectory data,this paper proposes an identification algorithm based on sliding window and LSTM automatic encoder to identify fishing boat operation mode.Firstly,the algorithm uses sliding window to extract the track features,which solves the problems of too long track length and big difference in the length of different track sequences.Then,LSTM automatic encoder is used to learn the time domain features and potential advanced features of the trajectory.Finally,softmax classifier is embedded in LSTM automatic encoder,and the loss function is jointly optimized to achieve the best classification effect.The AIS trajectory data of fishing boats in Zhejiang sea area verified the proposed fishing boat operation mode identification algorithm.The results show that the accuracy of the proposed method is 95.82%,and the algorithm can be used to assist the judgment of trawl and purse seine operation mode.