Deep neural networks for marine debris detection in sonar images
Valdenegro Toro, Matias Alejandro
MetadataShow full item record
Garbage and waste disposal is one of the biggest challenges currently faced by mankind. Proper waste disposal and recycling is a must in any sustainable community, and in many coastal areas there is signiﬁcant water pollution in the form of ﬂoating or submerged garbage. This is called marine debris. It is estimated that 6.4 million tonnes of marine debris enter water environments every year [McIlgorm et al. 2008, APEC Marine Resource Conservation WG], with 8 million items entering each day. An unknown fraction of this sinks to the bottom of water bodies. Submerged marine debris threatens marine life, and for shallow coastal areas, it can also threaten ﬁshing vessels [Iñiguez et al. 2016, Renewable and Sustainable Energy Reviews]. Submerged marine debris typically stays in the environment for a long time (20+ years), and consists of materials that can be recycled, such as metals, plastics, glass, etc. Many of these items should not be disposed in water bodies as this has a negative effect in the environment and human health. Encouraged by the advances in Computer Vision from the use Deep Learning, we propose the use of Deep Neural Networks (DNNs) to survey and detect marine debris in the bottom of water bodies (seaﬂoor, lake and river beds) from Forward-Looking Sonar (FLS) images. This thesis performs a comprehensive evaluation on the use of DNNs for the problem of marine debris detection in FLS images, as well as related problems such as image classiﬁcation, matching, and detection proposals. We do this in a dataset of 2069 FLS images that we captured with an ARIS Explorer 3000 sensor on marine debris objects lying in the ﬂoor of a small water tank. We had issues with the sensor in a real world underwater environment that motivated the use of a water tank. The objects we used to produce this dataset contain typical household marine debris and distractor marine objects (tires, hooks, valves, etc), divided in 10 classes plus a background class. Our results show that for the evaluated tasks, DNNs area superior technique than the corresponding state of the art. There are large gains particularly for the matching and detection proposal tasks. We also study the effect of sample complexity and object size in many tasks, which is valuable information for practitioners. We expect that our results will advance the objective of using Autonomous Underwater Vehicles to automatically survey, detect and collect marine debris from underwater environments.