Using neural networks to detect fracture patterns, in collaboration with Pierre Gentine (professor at Columbia University), Harold Li (data scientist at Lyft), Jonathan Kingslake (assistant professor at Columbia University), and Cameron Chen (data scientist at Google)
The availability of remote-sensing data is rapidly expanding. New strategies which take advantage of the most recent algorithms and approaches are needed to interpret large remote-sensing data sets. In order to identify fracture locations from satellite imagery across Antarctica, we develop a neural network model for fracture recognition. Ice fractures result in the collapse of Antartctica ice-shelves, which is expected to destabilize glacier flows into the ocean and accelerate glabal sea-level rise. The outcome of this research will provide a thorough map of fracture distribution across Antarctica, which is important for evaluating the impact of ice fractures on the future sea-level.
Fracture_detection.pdf : Summary of model performance (NN, DNN, CNN)
Fracture_UNet.ipynb: U-Net with large input shape (1000x1000) (DCNN)
Fractures Investigation_apply model_finepattern.ipynb : Comparison between differen model structure (CNN)
Fractures Investigation_apply model_finepattern_LargeInputShape.ipynb: Comparison between different input shape (CNN)