Scale-invariant image-recognition using convolutional neural networks and wavelet analysis

Heather K. Hardeman-Vooys, Matt A. D. McDonald, Michael P. Lamoureux

We begin with a discussion of machine learning and its applications to seismic inter-pretation. We provide a brief overview of convolutional neural networks as well as tree-structured wavelet transforms. We introduce a new wavelet transform called the inverted tree-structured wavelet transform which renders scale-invariance for image recognition. Also, we explain our method for processing data in real-time. Then we introduce a training set extracted from real data. We test the trained convolutional neural network on a portion of the training set to determine accuracy. Afterwards, we employ this trained convolutional neural network to identify events in microseismic data.