HPC-Europa3 visit: detecting floating objects in the sea using aerial images

29 March 2018

Dr Antonio Javier Gallego visited EPCC under the HPC Europa 3 Programme. He recounts his experiences in this blog article.

My name is Antonio Javier Gallego and I am from Alicante, Spain. I visited EPCC under the HPC-Europa3 programme hosted by Prof. Robert 'Bob' Fisher at The University of Edinburgh's School of Informatics. My research project consisted of the development of methods for the detection of floating objects in the sea from aerial images. In particular, two methods were proposed: one for the detection of oil spills and another for locating people in the sea (ie drowned, shipwrecked, or fallen overboard).

According to information obtained from the Spanish Maritime Safety Agency (SASEMAR), maritime traffic spills more than 20 million m3/year of both oil and other hydrocarbons into the waters of the European Union. The presence of oil slicks on the sea surface requires early detection to activate emergency protocols focused on controlling the environmental impact and ecological damage to be carried out. Detection and monitoring are usually performed manually using two kinds of sensors: images in a visible spectrum and Synthetic Aperture Radar (SAR). The presence of oil slicks on the surface of the sea requires early detection to activate emergency protocols focused on minimising the environmental impact and ecological damage. However, the locating of oil spills is a task that is usually performed manually.

The same problem occurs when trying to locate people in the sea. In Spain, small boats with immigrants arrive on the coast almost every day. Unfortunately, many of them are shipwrecked, and due to the waves and to the large area to explore, rescue missions are laborious and expensive. The maritime rescue service does not have any type of automatic system to help them in the search, therefore an application for this task is of great interest.

I used EPCC's Cirrus HPC platform with TensorFlow for the development and training of the necessary algorithms. The computational resources from this machine were mainly used to parallelise a grid-search process to explore several network configurations. As a result of this research I have written two scientific publications jointly with my host professor, which will be made publicly available along with the results obtained.

I really enjoyed my time in Edinburgh, since in addition to work I also had time for sightseeing. It is a really beautiful city, with lots of interesting things to see and shows to enjoy.