Actual-Scale Field Trial of Rapid Soil Classification with Computer Vision, complemented with Electrical Resistivity and Soil Strength

1. Home
2. Publications
3. Online Library
4. Actual-Scale Field Trial of Rapid Soil Classification with Computer Vision, complemented with Electrical Resistivity and Soil Strength

In Singapore, excess excavated soil from construction projects is re-purposed as material for infilling activities. Excavated soil is delivered to Staging Grounds (SGs) via tipper trucks, where they are categorized into either Good Earth or Soft Clay. Currently, soil classification is performed based on laboratory testing prior to excavation, and manual visual inspection of each truck upon arrival, which is highly human intensive. Heterogeneity of natural soils and possible mixing during excavation or loading process may also lead to a truck with differing soil types. This paper presents the prototype implementation of an innovative rapid soil classification system consisting of a computer vision technique with machine learning and complemented by geotechnical soil properties. The computer vision technique comprises soil image acquisition using an industrial-grade camera and decision-making with a trained Convolutional Neural Network (CNN) model. Three soil parameter determination at greater depth was used to complement computer vision: (i) apparent electrical resistivity using four electrodes arranged in Wenners array, (iii) moisture content using a time-domain reflectometer (TDR), and (iii) cone resistance using a modified cone penetrometer. A field trial was conducted using the on-site prototype setup on 493 soil samples and validated against the Particle Size Distribution (PSD) obtained from conventional laboratory testing. Results have shown that a high prediction accuracy of 85% was achieved. This non-destructive and instantaneous classification technique can be further expanded to other applications for sustainability in the construction industry.

9th International Congress on Environmental Geotechnics (ICEG2023)

Advances in Geoenvironmental Field Characterization