Autonomous Sea Turtle Detection Using Multi-beam Imaging Sonar: Toward Autonomous Tracking

Abstract

Tracking and logging marine animals are important for understanding them. Underwater robots such as ROVs and HOVs are used to observe marine animals manually. Recently, AUVs are also used to observe them. Toward wild-animal tracking by an AUV without attaching any tag to them, the authors propose an autonomous detection method with a multi-beam imaging sonar. Sea turtles are set as the first target. The method utilizes convolutional neural network (CNN) for detecting a turtle in multi-beam sonar imagery. For model training and evaluation, tank experiments and outdoor experiments were conducted, and sonar images of turtles were collected. CNN model was trained by images taken in the tank. The performance of the trained network was evaluated using the images obtained at the outdoor experiments.