Main content area

Measuring rock surface strength based on spectrograms with deep convolutional networks

Han, Shuai, Li, Heng, Li, Mingchao, Luo, Xiaochun
Computers & geosciences 2019 v.133 pp. 104312
computers, engineers, field methods, models, probability, rocks
One of the most widely accepted field methods used by geological engineers to measure rock surface strengths is by striking a rock with a geological hammer and using the emitted sound frequencies to determine strength. While the method is a convenient, it is also subjective. To this end, we propose a new method of measurement based on spectrograms using deep convolutional networks. The spectrograms collected through striking rocks with a geological hammer is the input variable to a deep learning model, which is the Inception-v3 model in this study and has a 93% classification accuracy. We then introduced a probability matrix and an error correction model to estimate the surface strength of rocks from the classification results. The experimental results show our method has high potential to underpin the implementation of efficient and objective measurements of rock surface strength in the field.