Smartphone-based image analysis method for estimating moisture content in subgrade filling

Abstract

Accurate and efficient estimation of moisture content during subgrade compaction is critical for ensuring compaction quality and structural integrity. Traditional measurement methods, however, are often lab-intensive, time-consuming, and costly. This study proposes a smartphone-based approach to estimate moisture content in subgrade filling, addressing these limitations. Laboratory experiments captured images of subgrade filling with varying moisture content, gradation and compaction levels. A color correction method based on a color calibration card was adopted to mitigate illumination-induced deviations, and an image segmentation technique was used to remove anomalous pixels such as cracks and pores. Feature importance was analyzed using five different methods to identify key variables, which were then used to train and evaluate multiple machine learning (ML) models. The proposed approach demonstrated high accuracy in estimating moisture content, offering a practical and scalable solution for subgrade compaction quality monitoring using smartphones.