Automatic detection of SEG-Y sampling format errors using machine learning
Sahana Vinayak, Ray Abma, Sergey B. Fomel · Second International Meeting for Applied Geoscience & Energy · 2022
SEG-Y datasets are occasionally misconverted when samples with an IEEE floating point format are converted as IBM float- ing point numbers or vise versa. This misconversion is often difficult to detect, but it will significantly increase the noise and corrupt the amplitudes. We show that automated format detec- tion with machine learning can be used to identify errors in floating-point number format conversions of a seismic dataset. A machine learning model was built using a deep artificial neu- ral network to classify the floating-point numbers. The model has a training accuracy of 96 percent and testing accuracy of 95.5 percent. Thus, the model is a practical method of iden- tifying the correct floating-point number format. This method can reduce incorrect conversions that may otherwise have gone undetected.