A novel approach to estimate sand particle-size using convolutional neural network with acoustic sensing

Abstract

Particle size of sand is one of the critical soil properties to estimate water flow-related phenomena (e.g., soil erodibility) and key soil properties such as hydraulic conductivity. This study proposed a new framework to classify particle size of sand using convolutional neural network (CNN) combined with ultrasound echo signals. The laboratory experiments were performed to construct the dataset of echo signals with different patterns as a function of median size of sand. The high accuracy of developed CNN model for classifying seven types of sand shown in this study implying the chance of using low-cost easy-to-measure ultrasound signals for monitoring median size of sand deposits. In addition, the accuracy of CNN models for the four scenarios shown in this study demonstrated the proposed framework in this study can be used to classify different sand type with low difference in median particle size. The developed CNN model in this study potentially can be used to monitor timedependent soil properties from ultrasound signals such as monitoring hydraulic conductivity of sand deposit.