Development of an algorithm based on deep learning for the classification of oceanic geophysical phenomena

Taipe Ramos L.N.; Sosa Callupe S.A.; Taipe Gallardo J.A.; Blas Morales O.E.; Palacios Bett J.S.; Rojas Vales N.C.; Rocha Calderon J.R.; Eulogio Saenz M. · Proceedings of the International Astronautical Congress, IAC (2022)

Resumen

SAR (Synthetic Aperture Radar) images are important for the mapping and monitoring of different geophysical phenomena, in this work we created a convolutional neural network called GeophysicalNet with the objective of extracting features from the 10 classes of phenomena present in the TenGeoP-SARwv database. The GeophysicalNet neural network is composed of non-sequential layers, residual layers, parallel layers and fully connected output layers with dropout. The neural network was trained with a fraction of the TenGeoP-SARwv database, we also trained several standards-based classifiers. We obtained the performance metrics of the GeophysicalNet neural network and traditional classifier models. The GeophysicalNet neural network proved to have higher accuracy on different performance metrics compared to other open-source neural network architectures. The proposed model can detect anomalies present in the ocean, with this open access information we have better control of maritime spaces and thus avoid environmental damage or natural disasters. Copyright

Instituciones: Universidad Nacional de Ingeniería