Submerged marine debris detection with autonomous underwater vehicles

Abstract

Garbage and debris produced by humans is commonly found in natural underwater environments such as oceans, lakes and rivers. Removal of submerged marine debris is required to prevent threats to marine and human life, and to maintain a sustainable environment. Detecting and mapping debris is hard due to unique difficulties of the underwater environment. In this paper we propose the use of Autonomous Underwater Vehicles to detect submerged marine debris from Forward-Looking Sonar (FLS) imagery. We train a Convolutional Neural Network to classify 96 χ 96 images and use this classifier as an object detector in a sliding window fashion. With this detector we can precisely separate debris from background seafloor. In a dataset of household marine debris captured with an ARIS Explorer 3000 FLS, we obtain 80.8% correct debris detections with a binary detector, and 70.8% with a multiclass detector. Our system can also generate detections for untrained objects, which indicates its good generalization capability. We believe our results point in a direction that Neural Networks combined with FLS form a powerful object detection system for small objects in underwater environments, which is ideal for marine debris.