Motion Behavior Recognition of Underwater Vehicle Based on YOLOv3

Abstract

Abstract As an important tool for human exploration and understanding of the ocean, underwater vehicle cannot achieve real-time cooperation due to the limitations of underwater acoustic communication. In order to realize the recognition and perception of the cooperative object behavior of underwater vehicle by vision, a new method is provided for the cooperation between underwater vehicle. Select YOLOv3 target detection algorithm, this paper to test the underwater vehicle motion behavior recognition, first of all, the collected experimental data set, using Label Image software training set and testing set of calibration, and modify the YOLOv3 classifier, change the output of the network dimension, optimization of network parameters and accelerate the convergence of the model, by analysing the experimental result shows that using YOLOv3 network, can realize the four directions to set an underwater vehicle motion behavior recognition, and ensure the accuracy and speed, the foundation for subsequent underwater vehicle based on visual collaboration.