Convolutional neural networks

Detection of hydrocarbon geologic "leads"
With Gabriel Leite
Detection of hydrocarbon geologic leads with convolutional neural networks

Process of patch generation from a simple binary image and a 3x3 patch model running with stride of 2. New image patches (subimages) are collectible to form a larger patch bank used for training of convolutional neural networks (CNNs) only if the area of white pixels (lead region) overlaps the running patch area by a proportion defined as "lead-pixel local area". In this example, two (blue and purple) of the three patches are non-collectible and one of them is collectible (red). Finally, the CNN is implemented for pattern recognition in seismic images.

Back to top