Combining convolutional and recurrent neural networks for ShakeMap prediction

Abstract

<p>     Taiwan has a high population density, and is located in the Circum-Pacific Belt, there are countless seismic hazard events.The Earthquake Early Warning (EEW) and Rapid Report Systems have become one of the most important disaster prevention systems to effectively reduce and prevent disasters caused by destructive earthquakes. Currently, according to the Central Weather Bureau's website, the EEW and repaid reports are produced and distributed 5 to 10 minutes after an earthquake, including the time of the earthquake, the location of the epicenter, the size and depth of the earthquake, and the fast estimated magnitude of the earthquake.<br />     In recent years, deep learning has developed rapidly, especially in the field of image recognition, including image classification, object detection, image generation, etc. This project is expected to use the Taiwan Strong Motion Instrument Program (TSMIP) station data provided by the Central Weather Bureau to generate the ShakeMap. We will use the multi-label classification method to mark the ShakeMap, and use the Recurrent Neural Network (RNN) to embed labels in the image, capture the dependency of labels and combine with a Convolution Neural Network (CNN) to predict the different labels in the shakeMap, which is expected to achieve the purpose of fast prediction of earthquake intensity in different places in the whole Taiwan island. </p>

Deep Representation Learning: Fundamentals, Technologies, Applications, and Open Challenges

“…There are myriad problems and use cases for the RNNs. Instances are: analysis and embedding of texts and medical reports to be combined with medical images [274], real-time denoising of medical video [275], classification of electroenephalogram (EEG) data [276], generating captions for images [28]- [30], [277], biomedical image segmentation [278], semantic segmentation of unstructured 3D point clouds [279], [280]. Other examples of RNN applications are predictive maintenance [281], prediction and classification of ICU outcomes [31], [282], [283].…”

Deep learning in ultrasound elastography imaging: A review

“…One of the major applications of RNNs is in image segmentation. 84 The structural characteristics of the RNN give it an inherent advantage for modeling sequence data. 21,85 RNNs were not widely utilized until recently due to difficulties in training them to capture long-term dependencies.…”