Deep learning-based digitization of raster well logging
Raster well-logging curves represent simulated data recorded during subsurface drilling operations,aiding geological experts in discerning specific lithology and strata based on the obtained curves.Presently,unsupervised computer vision tech-niques for interpreting digital curve algorithms still rely significantly on human intervention,exhibiting issues related to slower processing times and lower accuracy levels.This paper proposed a deep learning-based approach for the digitization of grid well-logging curves.The method leveraged an improved U-Net-inspired architecture as the foundational model,adapting the number of residual blocks during up and down-sampling processes to accommodate the high resolution of grid well-logging images.Addi-tionally,an attention-enhanced processing strategy was employed to balance the retention of critical signals when handling vari-ous input and output sizes.Semantic segmentation was applied to the grid images to reduce background interference and enhance digitization accuracy of curves.Experimental results demonstrated that the proposed architecture effectively classified and digi-tized curves compared to ground truth binary segmentation of multiple curve data,exhibiting high accuracy and performance in the digitization of raster well-logging images.
deep learningimage segmentationraster imagewell-log curve
NETL
NSTL
万方数据
