Abstract
Damage proxy maps are a type of computer-generated image that identifies damaged areas withing a target area. Currently most of the maps are created by the Jet Propulsion Laboratory through the ARIA program. The goal of this project is to recreate the quality of damage proxy maps by combining neural networks with publicly available data. To accomplish these two neural networks were utilized in combination of InSAR data to create maps to see which would create a more consistent result. The first experiment was a convolutional neural network architecture that was combined with phase maps created from the Alaska Satellite Facility (ASF) Hyp3 program. This experiment was ultimately dropped in favor of the second neural network as it required an extensive amount of data to train the algorithm and time that was incompatible with the duration of this experiment. The second experiment was a recurrent neural network architecture that was operated on coherence maps created using the ASF Hyp3 program. This algorithm required much less data and was able to generate damage proxy maps at a quality deemed acceptable for the purpose of this study. Further research into research into both neural networks could improve the quality of the results with future availability of higher quality public data.