A New Machine Learning Algorithm for Detection of Stray Clays

This paper discusses various machine learning algorithms utilized with Ground Penetrating Radar (GPR) for differentiating between clay seam number 414 and stray clays in potash mines to evaluate mine-room safety. Although different strategies have been used to find and recognize anomalies from GPR data through machine learning frameworks, but the frameworks which provide precise execution are still under investigation. In this research, we propose difference-of-gradients (DOG) method for detecting the presence of stray clays from the simulated and the actual GPR data, which is based on gated recurrent units (GRU) in neural networks. The combination of DOG and GRU methods assists to discover the accurate position of stray clays with time-series analysis. We optimize the strength of stray clay from clay seam number 414 by suppressing the stray clay signal strength from GPR data. A new window function in neural networks SmoothGrad which helps to filter the most of the anomalies effects on clay seam number 414 is utilized. This window length which tries to smooth GPR signal by supressing the noise, can be measured by adding the first and second gradients of GPR signal’s power. Therefore, my outcomes on the overall significances is to get the most accurate position of stray clays by differentiating the characteristics between the clay seam number 414 and the stray clays.

Name thesis:degree_name

Master of Applied Science (MASc)

Level thesis:degree_level

Master's

Discipline thesis:degree_discipline

Engineering - Electronic Systems

Grantor dc:publisher

Faculty of Graduate Studies and Research, University of Regina

Year dc:date.issued

2021

Author dc:creator

• Aysha, Umme

Advisor dc:contributor.advisor

• Paranjape, Raman

Committee member dc:contributor.committeemember

• Wang, Zhanle

Language dc:language.iso

en

DOI

https://doi.org/10.82465/4058

OAI identifier oai:identifier

oai:uregina.scholaris.ca:10294/14458