Online Program

WITHDRAWN: NumPy, Pandas, and PyTorch Libraries for Effective Computer Vision Implementation on Geophysical Data Sets (305359)

Keywords: machine learning, computer vision, Python, geophysical, earth science

Harnessing deep learning and computer vision to study the Earth has enabled various applications, such as monitoring natural hazards. From an implementation standpoint, the Python programming language is a useful and practical tool to utilize. More specifically, the NumPy, Pandas, and PyTorch libraries serve as key assets in developing neural networks that can analyze high-resolution imagery geophysical datasets, allowing for their application on future unseen data in real-world deployment. In this work, we delve into how NumPy and Pandas can aid in transforming and pre-processing the data into usable imagery for training the machine learning model, while explaining how PyTorch is the best Python library for actually developing the (convolutional) neural network, defining the loss function as the criterion for optimization, and training the network itself. We further focus on applications within these libraries that can help open "black box" models in order to discern the value of our results in relatively uninterpretable algorithms. We provide computer vision-based remote sensing of disaster damage assessment from satellite imagery as the primary dataset example in this poster.