12 October 2020 in Journal of Glaciology
In this study, our goal is to track internal ice layers on the Snow Radar data collected by NASA Operation IceBridge. We examine the application of deep learning methods on radar data gathered from polar regions. Artificial intelligence techniques have displayed impressive success in many practical fields. Deep neural networks owe their success to the availability of massive labeled data. However, in many real-world problems, even when a large dataset is available, deep learning methods have shown less success, due to causes such as lack of a large labeled dataset, presence of noise in the data or missing data. In our radar data, the presence of noise is one of the main obstacles in utilizing popular deep learning methods such as transfer learning. Our experiments show that if the neural network is trained to detect contours of objects in electro-optical imagery, it can only track a low percentage of contours in radar data. Fine-tuning and further training do not provide any better results. However, we show that selecting the right model and training it on the radar imagery from the start yields far better results.
19 November 2021 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Monitoring properties of ice sheets in polar regions is one of the main challenges in glaciology. There is a large amount of heterogeneous radar data from the polar regions that have been gathered through expensive missions. However, retrieving meaningful information from this large volume of data is still a great challenge. With the advancement of machine learning techniques in recent years, many scientists are eager to take advantage of these algorithms and techniques to explore and mine Arctic and Antarctic data. These advancements, however, have happened mainly in the area of supervised learning where the models are data hungry and require large amounts of annotated data. Generating simulated data can be an effective and inexpensive approach to provide large labeled datasets for training machine learning models. In this work, we explore two approaches to simulate arctic snow radar echogram images, namely a radar scattering physics based approach combined with some statistical measures and a purely data-driven approach based on a conditional generative adversarial network. Using several image comparison metrics, we analyze the utility of both methods for the purpose of simulating echograms. Our results show that the physics simulator generates images with good structural similarities, while the purely data-driven approach achieves better textural similarities for simulated image. Finally, we also show that by augmenting our real dataset by the simulated echograms, we can improve our deep learning model for tracking internal layers of snow.
1 April 2015 in Nonlinear Analysis