Developing Geological Structures Classification System using Deep Neural Network

The goal of every industry is to optimize production workflow, reduce cost in production, and make maximum profit. This as being the focus of the geoscience industries. We embrace the use of technology to maximize profits and reduce uncertainties. A good example is us moving from manually interpreting 2-D seismic data to using software like kingdom suite, Petrel E&P, etc. to interpret 2-D, 3-D, and 4-D data. This improvement has saved the oil and gas industries billions of dollars and as well helped to improve production.

Recently, Artificial Intelligence (AI) became the fourth industrial revolution that as began to drive every sector of our economy. The automobile industries, health sectors, real estates, and sections of the geoscience disciplines (Oil and Gas) have all focused on developing solutions tailored to the problem they are facing using this technology.

One of the major problem we face today as geoscientists is having to manually identify or classify the different rock samples or core data we acquired from the field. This process is time consuming and stressful. In this research, we looked into how we can leverage on AI to proffer solution to this problem. An Image recognition system to automate the classification of these rocks/core data was developed using convolutional neural network (CNN). To develop this system, a total of 2000 geologic images was used. 60% of the images were used as training data, 20% were used as validation data and 20% as testing data. To account for the underfitting error which was as a result of the limited amount of data available for training, data Augmentation analysis was carried out.

Currently, the system has a prediction succession rate of 65%. This implies that the system can successfully identify and classify 65% of the geological rock/core data feed into system. Further work is currently ongoing to ensure that a 99% success rate is achieved.