Convolutional Neural Network Based Target Recognition for Marine Search

Abstract

The key point of marine search and rescue is to find out and recognize the distress objects. At present, the visual search method is usually adopted to detect the ships in distress, and this method can only be used at good sea condition and visibility. In this paper, a new target detection and recognition system is proposed. The parameters of radar transmitter and echo graphics and the invariant moments of radar images are extracted as the system’s recognition features, and the system’s target classifier is based on Convolutional Neural Networks (CNN). The developed recognition classifier has been tested using three kinds of target Images, the target’s features are used as the inputs of trained CNN and the outputs of networks are target classification. Sea experimental results show that the proposed method is well-clustering and with high classified accuracy.