A target classification method for unmanned surface vehicle based on extreme learning machines

Abstract

In the process of autonomous navigation and obstacle avoidance of unmanned surface vehicles (USV), it is important for USVs to classify maritime targets correctly and effectively. In this paper, aiming at the recognition of surface targets for autonomous navigation of USVs, three kinds of targets are mainly considered, namely ships, buoys and islands. Visual sensors are installed on the USV to acquire visual images of maritime targets, and then the images are sent to the computer for automatic recognition. The invariant moments of three kinds of target images are extracted firstly, and target feature library will be built through image invariant moments, then an extreme learning machine (ELM)-based neural network is trained and then used to classify and recognise the sea targets. In addition, the sea targets are classified and analysed by AdaBoost-BP. The simulation results show that the ELM-based classification method proposed in this paper has a better performance for maritime targets.