Underwater Object Detection of AUV based on Sonar Simulator utilizing Noise Addition

Abstract

Object detection is one of necessary techniques for autonomous underwater vehicles (AUVs) to automate their missions. However, underwater object detection requires a large number of data images of target object. This paper proposes a method to generate highly reliable training images through sonar simulator and background noise templates. Sonar simulator has been developed to generate ideal images of target by modeling imaging mechanism of sonar sensor. To make the image realistic, background noise acquired in the blank water tank are added to the simulated images. Finally, the AUV could detect the target objects at sea using a convolutional neural network trained with the generated images without any field data which is difficult to obtain.