Grant: Alliance Grants

Ground Penetrating Radar (GPR) has been used to investigate the condition of roofs of mining rooms in Saskatchewan's potash mines for over a decade. This technology, used in conjunction with a mining borer allows the machine operator to obtain data about the characteristics of the roof in real-time as he is cutting into the potash seam. The idea is to allow the operator to recognize potentially unsafe conditions and make appropriate decisions about continuing the mining operation based on the GPR returns. Operation of the mining borer machine is intense work and it is difficult for the machine operator to concentrate simultaneously on both the boring and the GPR return signal analysis task. Additionally, assessment of the GPR return signal is a somewhat sophisticated task which is typically done by a trained geologist. Thus, it is not practical to expect the borer machine operator to be able to perform this analysis. In order to support the safe operation of the mining process, various analysis algorithms have been developed to automatically analyze the GPR returns. A confounding factor in these automatic analysis algorithms, called auto-picking algorithms, has been the complexity of the data. A particularly difficult issue has been the existence of stray clays in the region above the roof, which scatter and attenuate the GPR signal. These stray clays also attenuate GPR returns and this can lead to a reduced accuracy in the auto-picking algorithms. Today, machine learning algorithms have begun to dominate the field of feature detection in image and signal processing. These algorithms often outperform classical methods such as filtering, image transformation, and modeling for the detection of features in the data. However, machine learning has certain draw backs such as the need for large data bases of ground truth data which is highly representative of the actual environment. This research program is focused on developing and executing an appropriate machine learning strategy on the GPR return data in order to be able to process data accurately and at a sufficient rate so that it is practical to use this technology in the actual mine.