Exploring the Hydraulic Properties of Unsaturated Soil Using Deep Learning and Digital Imaging Measurement

The aim of this work is to improve the accuracy of the traditional Unsaturated Soil Hydraulic Property (USHP) analysis model. To this end, this work combines digital imaging measurement and Deep Learning technology to analyze the USHP. Firstly, the research status of basic characteristics of unsaturated soil and the application status of Deep Learning is analyzed. The second step examines the impact of soil specimens’ physical properties on their hydraulic properties. This includes acquiring hydraulic parameters and the Soil-Water Characteristic Curve (SWCC) based on a full-surface digital imaging measurement. Finally, an SHP model based on the Backpropagation Neural Network (BPNN) is implemented, trained, and validated. The results show that the proposed model’s predicted SWCC is consistent with the experimental phenomena of previous studies. Also, the proposed USHP BPNN model uses the Levenberg Marquardt (LM) algorithm to achieve the best results, with shorter computational time and less noise than other algorithms. Lastly, the results of the proposed model’s various parameters help determine that the optimal number of neurons in the hidden layer is ten. The proposed model’s error is more minor than several literature models thanks to considering the correlation between physical parameters like soil particle size and SHP. Overall, the BPNN is used to model the soil’s physical and hydraulic parameters and has shown excellent performance. The proposed USHP BPNN model helps streamline the traditional soil correlation coefficient estimation.