Enhancing Ground Penetrating Radar (GPR) Data Analysis Utilizing Machine Learning

Abstract

Ground Penetrating Radar is a non-destructive geophysical technique that utilizes radio waves to generate images of the Earth's subsurface to point out the location of buried evidence. In this paper, it is used to identify structures and types of seismic images of a real oil and gas field. This work employs GPR with 500MHz to permit the EMW to penetrate deep and to provide a good resolution for images generated. Gray-Level Co-Occurrence Matrix and Wavelet feature extractor approaches are mixed to extract 48 selected features. Subsequently, preprocessing techniques are utilized to improve GPR data analysis and interpretation, including refining data, imputing the missing values, normalizing all data, and splitting them into 70% for the training and 30% for the testing phases. Finally, various machine learning techniques are employed to classify the collected images using models like Decision Trees,agged trees, Naive Bayes, Artificial Neural Networks, Quadratic Discriminant Analysis, Support Vector Machines, and K-nearest neighbors. The performance metrics of all the machine learning approaches are worthy, and the proposed KNN can achieve an accuracy of 98.169%, 14 seconds of training time, and less than a few seconds of testing time.

Keywords

• Accuracy
• Classifiers
• Ground Penetrating Radar
• Seismic Images
• Real Oil and Gas Field