Fishing Vessel Type Recognition Based on Ship Position Data

Abstract

The type of fishing vessel operation is an important parameter for the fishing and management of fishery resources. The offshore motor fishing vessels in the East China Sea and the Yellow Sea were taken as the object of the research, and the BeiDou Vessel Monitoring System (VMS) position data from 2018 of these objects were used. The data is filtered and extracted according to the operating characteristics of the canvas stow net fishing vessel, and the trajectory map of the fishing vessel is drawn. On this basis, a fishing vessel classification method based on migration learning is proposed. This method uses VGG16 as the basic network, uses the parameters that have been trained on the ImageNet data set as the initial weights, and uses the preprocessing feature trajectory map as input of the network. Through training the model, the accuracy of the canvas stow net fishing boat and other fishing boats can be obtained. The experimental results show that the model classifies 4,974 canvas stow net voyages out of 54,120 effective voyages. The final accuracy rate was 91.8%, of which the recall rate of sailing net fishing boats was 91.9%, and the recall rate of other types of fishing boats was 91.8%. It provides a new solution for the identification of fishing vessel operation types.