Volcano Video Data Characterized and Classified Through Computer Vision and Machine Learning Algorithms

Video cameras are common at volcano observatories but due to the large data volume from continuous acquisition and time requirements for manual analysis, images are typically inspected after major eruptions. To use cameras as a monitoring tool, automated computer vision algorithms must distill raw images into relevant time series signals. Here we use a blob detection algorithm to highlight observable activity at Villarrica Volcano, Chile whose outputs characterize volcanic activity but are orders of magnitude smaller than the original imagery. These outputs are then fed into a supervised artificial neural network to classify the observable activity into five classes.