HPC-Based Navigation System for Marine Litter Hunting

Presentation of the problem and objective of the experiment

Protecting seas and oceans against the litter is becoming a global concern and there is a growing need worldwide for more efficient, clean and autonomous technologies to identify and collect marine detritus, especially plastics, in a systematic and repetitive way. GTS is an agile and innovative start-up that operates in the field of environmental protection and blue growth economy. The use of HPC makes it possible to tackle a computational problem that GTS met during its service for recovery plastic litter in sea: optimizing the plastic litter recovery strategy forecasting the position of hundreds of detritus floating in the sea with suitable accuracy in space and time.

Short description of the experiment

The proposed HPC experiment is fundamental to drive into the next phase the collaboration of the unmanned systems as it requires >250.000 hours of deep learning which is impossible under conventional computational systems but possible thanks to CINECA. The limited forecasting capability of the future position of detritus thus is limiting the efficiency of recovery of the whole system. The HPC experiment aims to overcome this limitation and targets to improve the current Deep Learning approach to 1. Identify and classify marine litters in terms of dimensions and materials (PET, PPT, Biological); 2. Predict the possible trajectories of classified waste over a longer time; 3. Search the “best” trajectory to collect as much waste as possible under constraints. 



Partners have completed the data acquisition part with aerial drone shots both for the images of the waste to be classified (bottles, caps, bags, glasses, etc ...) and with respect to the real trajectories of typical waste such as bottles (full, empty and a half). At the same time, acquisitions of marine litter movement with sensorised buoys were also carried out and the first oceanographic simulations were started to generate other dragging data useful for training.
Partners are in the test phase for the neural networks identified for the recognition of marine litter and the prediction of trajectories. They expect to be able to find those that generate the best accuracies by the end of the year . The synergy with the entire consortium is proceeding at its best and the production of the progress reports is proceeding as planned.



Organisations involved: 

End User: Green Tech Solution SRL
Domain Experts: Università degli Studi di Napoli Parthenope and BI-REX - Big Data Innovation & Research Excellence
HPC Provider: CINECA 


Partners CINECA and BI-REX are part of the NCC Italy.