Field Trial on Rapid Soil Classification Using Computer Vision

This paper presents a rapid soil classification method using computer vision to improve the productivity of on-site classification of excavated soil. In Singapore, thousands of truckloads of excavated soil are transported from construction sites to transfer hubs (known as "staging grounds") daily. These soils, broadly classified into "Good Earth" and "Soft Clay," are reused for various applications in land reclamation works, with or without further treatment. An accurate classification is needed for better re-utilization of these material for a sustainable built environment. However, the current classification methods available are expensive, time-consuming and/or labor-intensive, while visual inspection is subjective and prone to human error. A proof-of-concept study of the rapid classification method using computer vision was conducted at a staging ground. Results showed that the optimized artificial neural network model, using three image textural features (Contrast, Correlation, and Entropy) can classify soils into "Good Earth" or "Soft Clay" in less than three minutes, with an accuracy of at least 85%.