Seismic image recognition tool via artificial neural network

In oil and gas exploration, seismic images are processed to identify the existence of potential reservoir by classifying the seismic image into different sections. These sections, also known as objects, are made up of different patterns portraying the structure of subsurface. This study aims to develop an artificial neural network to recognize the objects of channel and fault in seismic images. Three neural networks employing tan-sigmoid, log-sigmoid and purelin transfer function were created respectively. Gray Level Cooccurrence Matrix (GLCM) textual feature is used as image features in our dataset. The accuracy of the developed neural network in recognizing channel and fault in seismic images were measured. This preliminary study reveals that the feedforward neural network with transfer function of tan-sigmoid has the best performance in classifying the objects in our case. It is then used to develop an automated tool as our prototype system to facilitate seismic object recognition. It is observed that the prototype system can serve as a good tool for undergraduate students to learn about channel and fault recognition with minimal guidance from the experts.