Conditional Curve Segmentation for Raster Digitization
M.S. Shakeel · 2024
Summary Raster well log files encapsulate invaluable subsurface information derived from drilling operations, representing fundamental rock properties such as porosity, density, gamma ray measurements, among others, typically recorded at a resolution of 2 datapoints per foot. These datasets trace their origins to a time predating digital computing when measurements were manually recorded on graph paper during well drilling operations. Over the years, these analog records have been preserved as scanned PDFs, presenting a wealth of historical subsurface data vital for contemporary geoscientific analysis. The process of digitizing these logs into csv format involves segmenting out well logs curves from plot segments. Contrary to typical segmentation problem this problem involves taking the line style as input along with the plot image and segment out the corresponding curve. We call this conditional segmentation as the segmentation of a plot segment depends on the curve we want to extract (the condition). In this paper we present to you an innovative way of solving the above conditional segmentation problem using trivial semantic segmentation models.