Intelligent recognition and analysis method of rock lithology classification based on coupled rock images and hammering audios

In the process of field geological exploration, geological engineers generally make a preliminary judgment on the lithology classification through the information of rock surface texture, color and hammering audios. Based on experts' experience and intelligent mode, a deep learning and intelligent recognition analysis method for rock lithology classification is proposed by using coupled rock images and hammering audios. Firstly, the Transfer learning method based on Inception-V3 is used to carry out deep learning and training on collected 6 types of rock images. Then, measuring the rock strength indicated as the average value of the rebound index by a rebounder and obtaining hammering audio segments by the threshold method, an SVM (Support Vector Machine) regression model of waveform and intensity is established to predict the rock surface strength. Finally, rock classification is intelligently recognized by coupling rock images recognition model with rock audio intensity regression model. The accuracy of rock classification is increased from 83.5%(only using simple images recognition model) to 90.5%. The coupling model can not only effectively identify rock lithology classification but also preliminarily give the rock surface strength, which provides a new auxiliary method for field engineering geological survey and is beneficial to improve the working efficiency of initial field exploration. © 2020, Science Press. All right reserved.