A comparative study of Deep Learning architectures for Classification of Natural and Human-made Sea Events in SAR images

• DOI: 10.1007/s44163-022-00017-5
• Publication Date: 2022-02-02T11:06:06Z
• Authors: William Ramirez; Pedro Achanccaray; Marco Aurelio Pac...
• Keywords: Deep learning; QA75.5-76.95; 02 engineering and technology; Remote sensing; 13. Climate action; Electronic computers. Computer science; Maritime events classification; Computational linguistics. Natural language processing; 0202 electrical engineering, electronic engineering, information engineering; Convolutional neural networks; Sentinel-1A; 14. Life underwater; P98-98.5

Abstract Sea monitoring is essential for a better understanding of its dynamics and to measure the impact human activities. In this context, remote sensing plays an important role by providing satellite imagery every day, even in critical climate conditions, detection sea events with potential risk environment. The present work proposes comparative study Deep Learning architectures classification natural man-made using SAR imagery. evaluated comprises models based on convolutional networks, inception blocks, attention modules. Two datasets are employed purpose: first one encompasses series (geophysical phenomena), while second describes real oil spill scenario Gulf Mexico from 2018 2021. As result, through experimental analysis, it demonstrated how Xception Attention sampling obtained highest performance metrics, presenting Recall values 94.2% 87.4% human-made events, respectively.