Research On The Behavior Identification Method Of Fishing Vessels Based On Deep Learning

Ship Automatic Identification System(AIS)is a new type of navigation aid system.It was originally used to ensure the safe navigation of ships.With the advent of the era of big data,AIS data has been widely used.In this paper,AIS data is used to identify the fishing behavior of fishing vessels.Through the research on the behavior patterns of fishing vessels,the relevant departments can understand the distribution of hot fishing areas and the evolution of fishing grounds macroscropically,so as to formulate a reasonable fishery plan and ensure the healthy and sustainable development of fishery economy.The main contents of the paper are as follows:(1)the trajectory data of fishing vessel is extracted from AIS data as the initial data set,then the linear interpolation method is used to fill in the missing data,and the abnormal data of position and velocity are eliminated for data preprocessing.Two traditional machine learning methods including the Gaussian Mixture Models(GMM)and Hidden Markov Models(HMM)are used to identify the fishing behavior of fishing vessels.The input of two methods is only the single feature of vessel velocity.In GMM method,the parameters of Gauss distribution of speed are obtained by EM algorithm.The interval [? ± ?] composed of the mean value ? and the standard deviation ? corresponding to the medium speed is regarded as the confidence interval of the velocity of a fishing vessel to identify its fishing behavior.This method requires that the velocity feature satisfies the Gaussian distribution,otherwise the effect will become worse.In the HMM method,the fishing behavior of fishing vessels is identified by determining the transfer matrix of different behavior states,which is not affected by the velocity distribution.Neither GMM nor HMM makes full use of otherfield information in AIS data.(2)A fishing vessel behavior recognition method based on Long Short-Term Memory(LSTM)is constructed,which identifies fishing activities from the features of time series.The input,output and structure of the LSTM model are determined,and the AIS data field information is fully utilized.The model input includes six features:velocity,longitude,latitude,heading orientation,length and width of the vessel.The parameters of the model are selected through experiments,and the accuracy,precision,recall and F1-score are taken as the evaluation indexes of the experimental results.The effects of Timestep values and data input on the experimental results are discussed and compared with other methods.(3)A fishing vessel behavior recognition method based on Convolutional Neural Networks(CNN)is constructed.The input features are longitude,latitude,speed,heading,length and width.This method uses two layers of convolution operations,one layer convolution is used to learn the local features of the data,and the other layer convolution is used to increase the nonlinearity of the network.The feasibility of this method is verified through experiments.The paper also constructs a new method to identify the behavior of fishing vessels based on Convolutional LSTM Network(ConvLSTM).The method first learns the local features of the data through CNN,then inputting the learned features into LSTM to learn the temporal correlation of the features.Finally,the advantages of this method in the identification of fishing vessel behavior are verified by experiments,which are compared with the single LSTM and CNN methods.Experimental results show that compared with the traditional GMM and HMM machine learning methods,the depth learning methods LSTM,CNN and ConvLSTM are more effective.The performance of LSTM is better than that of CNN,and the processing speed of CNN is much faster than that of LSTM.ConvLSTM has the best recognition effect,which combines the advantages of CNN and LSTM,with higher recognition rate and faster convergence speed.Compared with LSTM alone,ConvLSTM greatly reduces the computation time,and its accuracy,precision,recall and F1-score are 0.9895,0.9701,0.9893 and 0.9796 respectively,which shows that this method has great application value.This provides other alternative techniques for identifying fishing behavior technology of fishing vessels,which is conducive to monitoring the status of fishing vessels,discovering fishing grounds and formulatingfishery resource planning,so as to ensure the healthy and sustainable development of fisheries.